Evaluating the Effect of Noisy Thermal Images On the Detection of Early Breast Cancer Using Deep Learning

Evaluating the Effect of Noisy Thermal Images On the Detection of Early Breast Cancer Using Deep Learning | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Evaluating the Effect of Noisy Thermal Images On the Detection of Early Breast Cancer Using Deep Learning Mohammed Abdulla Al Husaini, Mohamed Hadi Habaebi, Elfatih A.A. Elsheikh, and 3 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-3978583/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Breast cancer remains a leading cause of mortality among women globally. There were techniques that have been developed to enhance early detection, among which thermal imaging has emerged as a promising modality capable of identifying potential signs of breast cancer in its early stages. In addition, Thermal images provide valuable pixel-level information by capturing temperature variations between healthy and cancerous tissues. However, the susceptibility of these thermal images to noise poses a challenge to the diagnostic accuracy in early stages. This research aims to assess the influence of various types of noise on performance of recently developed different deep learning models designed for early breast cancer detection. In addition, a comprehensive analysis was conducted using a substantial database to assess the impact of noise on the models' efficacy. Also, encompasses different categories of noise characterized by distinct mean and variance values ranging from 0.01 to 0.09. The findings reveal that the introduction of different types of noise, albeit within a small range of mean and variance values, adversely affects the performance of deep learning models. It shows that these filters play a pivotal role in enhancing the accuracy of classification. Moreover, the results show that salt and pepper noise, varied between 0.1 and 0.3, significantly impacted the accuracy of inception MV4, reducing it from 100–51.58%, without adding filters in pre-processing. Additionally, the introduction of variance in multiplicative noise from 0.2 to 0.8, demonstrated an effect on classification accuracy only at noise levels of 0.7 (89%) and 0.8 (43%). Moreover, the results show that performance metrics for proposed method were accuracy of 99.82%, sensitivity of 0.996, specificity of 1, precision of 1, NPV of 0.997, FNR of 0.004, LRN of 0.004, AUC of 0.998, EER of 0.002, and F1 score of 0.998, but FPR of 0. In conclusion, findings underscore the significance of refining both noise mitigation strategies and preprocessing techniques to advance reliability and accuracy of thermal imaging as a diagnostic tool in breast cancer detection in early stages. Breast Cancer Thermal Image Gaussian Noise Deep Learning Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 1. Introduction According to the World Health Organization, statistics indicate an increase in the incidence of breast cancer. The statistics mentioned in 2020 indicate that there are 2.088 million new cases of breast cancer, and the most important reason is late detection [ 1 ]. Machine learning and deep learning have emerged as transformative technologies in the field of medical sciences, revolutionizing various aspects of healthcare, including disease diagnosis in early stages and treatment [ 2 ]. In addition, these advanced techniques have played a pivotal role in improving accuracy and efficiency in the realm of breast cancer detection [ 3 ][ 4 ][ 5 ]. Moreover, with vast amounts of accurate medical database available, machine learning algorithms can analyse intricate patterns and anomalies within mammograms, ultrasound images, and pathology slides, aiding in the early detection of breast cancer. Deep learning models have demonstrated remarkable capabilities in image recognition and classification tasks, enabling healthcare professionals to identify suspicious lesions with high precision. Moreover, these technologies have the potential to assist radiologists and pathologists by providing them with decision support tools, thereby enhancing diagnostic accuracy and reducing the likelihood of human error [ 6 ]. Recently, thermal imaging technology has been utilized to detect breast cancer using deep learning. Therefore, it is a safe, harmless, and promising technique for early detection [ 7 ] This technology has become a home diagnostic tool that enables patients to track tumor status periodically [ 8 ]. Therefore, it is subject to Noise in some cases that may not comply with thermal imaging procedures. Therefore, one of the factors affecting the classification of thermal images in deep learning is location and size of tumor in breast and the presence of Noise in thermal images. This study will explain the im-pact of Gaussian Noise on thermal images and on databases. Image noise, which is undesired information within an image, can manifest at various stages like image capture, transmission, or processing. In addition, a comprehensive understanding of the noise's characteristics is essential to effectively eliminate noise from a noisy image. On the other hand, attempts at noise removal may lead to image blurring [ 9 ]. The study in [ 10 ] explores the diverse effects of noise types, such as Gaussian, Salt & Pepper, Speckle, and Poison, on thermal image features—specifically Histogram, Texture, and Trans-form-based features. Evaluating both pre- and post-noise application, Signal-to-Noise Ratio and Mean Absolute Error serve as quality criteria and Results underscore the substantial influence of the Mode feature in thermal images compared to regular im-ages. The fundamental idea is to enhance the conventional Laplacian of Gaussian filter by incorporating interval analysis to account for intensity uncertainties. Also, experimental evaluations were conducted on MIAS, a widely used breast cancer database, for medical image segmentation. Finally, system's performance was assessed in comparison to Prewitt, LoG, and Canny filters using the Peak Signal-to-Noise Ratio (PSNR) [ 11 ]. On the other hand, images captured from a Mammogram might exhibit noise caused by variations in lighting and sensor inaccuracies. Cancer’s noise can be effectively removed without compromising the image's boundaries and small details, accurate diagnoses of breast cancer can be facilitated through imaging technology [ 12 ]. 1.1 Related works The study [ 13 ] referred to the use of the numerical simulation system for the location of tumors in the breast on different sites and the addition of thermal cooling. Gaussian Noises were added to thermal images. In addition, only four tumor sites were placed in the breast at different positions. The obtained results claim that the time-domain phase approach shows improved detection capabilities over those of raw heat. The study conducted by the researcher [ 14 ] proposed a new model of Deep Convolutional Neural Network called SPARs. This model has been proposed for low-dimensional deep heat extraction. In addition, 208 clinical breast cancer thermal images were used with the DMR IR database. Moreover, Gaussian Noise was added at a rate of 3 to 20% on the thermal images, but after encoding the results showed a slight decrease in the accuracy of the thermal images. SPAER demonstrated great robustness when tested for additive Gaussian Noise conditions (3–20% Noise), as assessed by its signal-to-noise ratio (SNR). The results indicate a high performance of SPAER to maintain thermal heterogeneity, and it can be used as an in vivo non-invasive tool that aids CBE in the early detection of breast cancer. In [ 15 ] used the Deep Convolutional Neural Network and BIOS optimization algorithm. In addition, thermal images used were 3895, divided as follows: 3098 healthy and 757 cancers. also referred to the processing of images with added Noise. The results indicate an increase in the accuracy rate from 97.91 to 98.95%. The researcher [ 16 ] refers to processing thermal images and the removal of Noise by using the soft wavelet threshold before color segmentation. In addition, the proposed method was applied to 50 thermal images after applying pre-treatment. Clinical thermal images were used by a 640 x 480 pixels thermal camera. The results indicate that carefully designed color maps visually improve the thermal image, and improve pest detection and interpretation without using any algorithms or deep learning. The study presented by [ 17 ] utilized Cascade Deep Convolutional Neural Network to classify thermal images of breast cancer. The database consists of 900 thermal images. Gaussian Noise has been added to increase the database of thermal images. In addition, Deep Convolutional Neural Network was trained on thermal images for automatic classification. CNN was trained by setting 50 epochs and learning rate of 0.5e-3. The results indicate that accuracy reached 92%. The researchers [ 18 ] referred to utilize of the deep convolutional neural network BreaCNet model, which is a modified deep convolutional neural network Shuffle Net model. Also, it can extract around 6 million features. In addition, 1302 thermal images were used for training 90% and testing 10% which were collected from DMR IR database. Moreover, Sobel kernel was used to filter the edges in the region of interest (ROI) and convert color thermal images into grayscale. In addition, the modified model deep convolutional neural network was tuned learning 1e -3 with 75 epoch and SGDM optimization method was used. Results indicated the accuracy of the detection reached 100% compared to Mobilnet, which reached 98%. Furthermore, a Gaussian filter has been added and smartphone app has been created with size 22MB. In [ 19 ] a new breast cancer detection method, DBC-4D U-Net-DITI, utilizing digital infrared thermal imaging. Also, it employs Altered Phase Preserving Dynamic Range Compression (APPDRC) for preprocessing, optimizes 4D U-Net weights using Glow-worm Swarm Optimization Algorithm (GSOA) for segmentation, reduce speckle noise, and employs a Binarized Spiking Neural Network (BSNN) for pathology stage classification. In conclusion, results indicate in comparison to existing methods like DBC-CSSA-DITI, DBC-MPA-DITI, and DBC-CNN-DITI, the proposed approach demonstrates substantial improvements with a 39.01%, 28.34%, and 37.45% accuracy boost and a 17.12%, 24.12%, and 32.07% precision enhancement, respectively. It is clear from the previous studies and Table 1 that Gaussian Noise was used in these studies. However, these studies lack clarity of noise data. Therefore, the proposed approach introduces 3 different types of noise with different values to a database consisting of thermal images. In addition, the effect of an added set of noise values on the classification capabilities of Deep Convolutional Neural Network Inception MV4 is studied. In conclusion, filters were added to mitigate the effect of noise from thermal images before classification. 1.2 Motivation Breast cancer is a significant health concern worldwide, and early detection con-tributes to improving patient outcomes. Thermal imaging, as non-invasive, shows promise in detecting breast cancer in early stages. However, thermal images are susceptible to different types of noises, such as Gaussian noise, speckle noise, salt and pepper noise, and Poisson noise which will affect image quality. Therefore, under-standing the effects of different types of noises on thermal images of breast cancer are essential for developing robust and reliable deep learning algorithms for accurate detection and diagnosis. 1.3 Contribution The evaluation of Gaussian noise effects in thermal images of breast cancer and its impact on deep learning methods have an important contribution: (i) The study provides a thorough evaluation of the effects of different types of noises on thermal images specifically related to breast cancer. By studying the effect of noise on image quality, important features, and overall diagnostic performance, enhances our understanding of the challenges associated with noisy thermal images. (ii) The research examines the performance of deep learning algorithms, such as convolutional neural networks (CNNs), in the presence of Gaussian noise. By quantifying the degradation in accuracy, sensitivity, and specificity caused by noise, it provides insights into the limitations of existing deep learning models and highlights the need for noise-aware techniques. (iii) Based on the evaluation results, the study investigates and proposes effective strategies to mitigate the adverse effects of different types of noise on thermal images. This could include preprocessing techniques, noise reduction algorithms, or modified deep learning architectures that are robust to noise, ultimately improving the accuracy and reliability of breast cancer detection systems. (iv) The findings and recommendations from this research have practical implications for medical professionals, researchers, and developers working on thermal imaging-based breast cancer diagnosis. The study can guide the development of noise-robust deep learning models, contribute to the design of optimized imaging protocols, and assist in the implementation of quality control measures for thermal imaging devices. By addressing the motivation and making these significant contributions, the evaluation of Gaussian noise effects in thermal images of breast cancer and deep learning advances our understanding of the challenges associated with noisy thermal images and helps pave the way for improved early detection and diagnosis of breast cancer. 2. Materials and Methods In this section, we will introduce the deep learning framework used in this paper. Namely, inception v3, inception v4, and modified inception v4 models [ 20 ] are introduced below. 2.1 Inception v3 Inception-v3 is a convolutional neural network (CNN) architecture that has been created in 2015 by Google researchers. It is part of the Inception family of CNN models and is designed for image recognition and classification tasks. Inception-v3 is characterized by its deep architecture, consisting of 48 layers in total. It uses a combination of different size of filters in convolutional layers, max pooling, average pooling layers, and fully connected layers. The architecture is based on the idea of "inception modules," which are building blocks that allow for efficient multi-scale feature extraction. The main innovation of Inception-v3 lies in its inception modules. These modules use various filter sizes (1x1, 3x3, 5x5) to capture information at different spatial scales and process them in parallel. This helps the network capture both fine-grained details and high-level features. To reduce the computational complexity of the network, Incep-tion-v3 uses 1x1 convolutions as dimensionality reduction layers. These layers reduce the number of input channels before applying larger convolutional filters, thereby im-proving efficiency without significant loss of information. Inception-v3 includes auxiliary classifiers at intermediate layers. These classifiers help combat the vanishing gradient problem during training and provide regularization. They also serve as additional sources of gradients during backpropagation, aiding in the overall training process. Inception-v3 is typically pre-trained on large-scale datasets like ImageNet, which contains millions of labelled images from various categories. After pre-training, the model can be fine-tuned on smaller datasets for specific image recognition tasks, adapting its learned features to the new dataset. Inception-v3 achieves impressive performance on image classification benchmarks. In the ImageNet Large Scale Visual Recognition Challenge (ILSVRC) 2015, it achieved a top-5 error rate of 3.46%, surpassing its predecessor models and demonstrating state-of-the-art accuracy at the time. Inception-v3 and its subsequent versions have been widely adopted and used as a backbone architecture in many computer vision applications, including object recognition, image segmentation, and transfer learning tasks. Its design principles have also influenced the development of other CNN architectures, such as Inception-v4 and Inception-ResNet [ 21 ]. 2.2 Inception v4 Inception-v4 is an advanced convolutional neural network architecture that im-proves upon previous versions of the Inception family. It simplifies the overall structure, introduces a stem layer, and incorporates a greater number of inception modules compared to Inception-v3. The architecture of Inception-v4 is characterized by a more uniform and simplified design. The network consists of a comprehensive planner and stem configuration, along with 4 inception A layers, 7 inception B layers, and 3 inception C layers [ 22 ]. 2.3 Inception mv4 An examination was conducted to compare "inception B" in Inception V4 to the updated version in MV4. The modifications made were limited to "inception B" as shown in Fig. 1 . First, a convolutional layer was added beneath the average pooling layer. The number of filters was increased from 128 to 256 to maintain the number of features extracted. Then, two parallel convolutional layers were added under the layer with 192 filters, both with the same size and number of filters. After that, the remaining layers in inception B were removed to keep the number of extracted features. Finally, Inception B has 7 groups, each with a set of layers. All 7 groups were changed with the same adjustments mentioned before. For further details on the framework, readers are advised to consult [ 20 ]. 2.4 DMR IR Database for Mastology Research with Infrared Image The patient must abstain from hot liquid, activity, and applying any creams to their breasts and underarms at least 2 hours before the exam. Moreover, the exam room should be kept at a temperature between 20–22°C. Also, Patient Preparation: In the exam room, the patient should remove any jewellery or accessories that could affect the thermal image. Additionally, the camera should be positioned 1 meter from the patient. The acquisition should be dynamic with the patient facing the camera. IR images are taken using a FLIR SC620 with a sensitivity of less than 0.04°C and a temperature range of -40°C to 500°C. It can be accessed through an online interface ( http://visual.ic.uff.br/dmi ) that allows for easy management and retrieval of information from breast exams and patient clinical data. 2.5 Speckle noise model A speckle pattern that appears in polarized monochromatic light can be viewed as being caused by a classical random walk in the complex plane [ 23 ]. Speckle noise is a type of noise that appears in images as a rough and grainy pattern resembling salt and pepper noise. It has the greatest impact on thermal image features, while Salt & Pepper noise has the lower impact [ 10 ]. This noise referred to is a multiplicative noise with a granular form. The presence of speckle noise in images is unwanted as it degrades the thermal image quality by impacting the edges and local details between different organs, which are crucial for diagnostic purposes [ 24 ]. Speckle noise in the medical literature is known as "texture" and might hold valuable diagnostic data. The ideal level of speckle smoothing is largely influenced by the expert's expertise and the intended use. In automatic segmentation, preserving the crispness of boundaries between several image areas is commonly selected while reducing the speckled texture [ 25 ]. 2.6 Gaussian noise model Gaussian noise, also known as additive white Gaussian noise (AWGN), is a common type of random noise that occurs in many imaging and signal processing applications. It is characterized by random variations in pixel intensities that follow a Gaussian distribution. It is often introduced during image acquisition, transmission, or processing due to various factors like electronic noise, quantization errors, or sensor imperfections. The noise appears as a random fluctuation around the true pixel value, resulting in a smooth distribution of noise values across the image. Moreover, the properties of Gaussian noise make it an important noise model in many theoretical and practical ap-plications. In addition, the Gaussian distribution is symmetric and bell-shaped, with most noise values concentrated around the mean value and fewer values occurring at the extremes. However, the noise values are statistically independent at each pixel, and their distribution is characterized by the mean and standard deviation parameters. The choice of denoising technique depends on the noise characteristics, the desired level of noise reduction, and the trade-off between noise removal and preservation of image details [ 26 ]. 2.7 Salt and pepper noise Salt and pepper noise is also known as impulse noise, and it is a common type of image noise that can affect the quality and clarity of digital thermal images. It is characterized by the occurrence of randomly scattered white and black pixels, resembling grains of salt and pepper. It can occur due to various factors such as transmission errors, faulty sensors, or electronic interference during the image acquisition or processing stages. It typically manifests as isolated bright white pixels (salt) and dark black pixels (pepper) randomly scattered throughout the image. The presence of salt and pepper noise can significantly affect image analysis and processing tasks, as it introduces un-wanted artifacts and distorts the overall appearance of the image. Therefore, it is crucial to employ denoising techniques to mitigate the effects of this noise and restore image quality. When selecting a denoising method, it is important to consider the trade-off between noise removal and preservation of important image features. Different methods may yield varying results depending on the characteristics of the noise and the desired image quality [ 27 ]. 2.8 Poisson noise Poisson noise is a type of random variation that commonly occurs in images captured under low-light conditions or when using imaging devices with photon-counting sensors, like medical imaging devices. Poisson noise is caused by the random nature of photon arrival at the sensor or detector. In addition, it manifests as a variation in the number of photons detected at each pixel, resulting in intensity fluctuations in the im-age. The Poisson distribution models the probability of observing a certain number of events (photons) in each interval of time or space. In digital images, Poisson noise appears as random variations in pixel intensities, with brighter areas having a higher probability of more photons and darker areas having a lower probability. This noise can degrade image quality by reducing contrast, introducing unwanted variations, and affecting image analysis and processing tasks. Moreover, it is inherently different from other types of noise, such as Gaussian noise, due to its statistical properties. Therefore, specialized denoising techniques are required to effectively address this specific type of noise [ 28 ]. 2.9 Experiment setup The experiments conducted to detect breast cancer at an early stage were by adding Gaussian Noises of various proportions. The results obtained were compared with the results in previous studies. In addition, the experiments were divided into four stages. The first stage is to use Gaussian Noise in the thermal images database only. The second stage is to use the original database without adding noise which is used to verify the modified Deep Convolutional Neural Network inception MV4. The third stage is the merging of original thermal images from the DMR IR database with thermal images with added Gaussian Noise. The fourth stage is to add Noise filters before classifying thermal images in a modified Deep Convolutional Neural Network inception MV4. As for the devices used in the experiments: a computer with high specifications, 32 GB RAM, Core i7, and a storage capacity of 1 TB. In addition, MATLAB version 2020a software and graphics processor unit. Moreover, thermal images from the DMR IR database were utilized, with a total of 1800 thermal images categorized as healthy (800 thermal images) and unhealthy (1000 thermal images). The deep Convolutional Neural Network inception MV4 model was used because it has high accuracy compared to other neural networks as shown in Fig. 1 [ 20 ]. 3. Results and Discussion Figure 1 indicates that the change in Gaussian Noise values affects to classification quality of the Deep Convolutional Neural Network, as healthy thermal images were classified as diseased. Gaussian Noise was added at the following values 0.05 and 0.02 in all thermal images of the database and the new database was saved. In addition, Deep Convolutional Neural Network Inception MV4 was set to a learning rate of 0.0001, SGDM optimization method and training 10 epochs see Figure 2. As for the second stage, Deep Convolutional Neural Network was trained using the same settings, but without adding Gaussian Noise. The third stage was the training of Deep Convolutional Neural Network on the same settings, but by adding actual thermal images with added Gaussian Noise. The following graph shows the amount of change in Noise compared to the accuracy of classification in each of DMR IR and DMR IR databases with Noise and with Gaussian Noise. Where the results indicate that the database without Noise has best solution for classification. As for database with Noise in any of the thermal images, filters are added during image processing and before entering Deep Convolutional Neural Network. Figure 3 is clear from the comparison between thermal image 1, thermal image 2, and thermal image 3 constitutes the difference between the addition of Gaussian Noise and the actual thermal image. Moreover, it is clear from Figure 4 gradient in addition to Gaussian Noise and the extent of its impact on the accuracy of classification. Table 2 shows that the analysis reveals a mean noise value, a standard deviation of noise, and a noise value in dB from Figure 4. These parameters impact on the classification accuracy in thermal images. Fourth stage: thermal image processing filters built before being inserted into Deep Convolutional Neural Network. Detection accuracy was calculated and found to highlight the effect of filters on the classification using Deep Convolutional Neural Network. The results indicate that the accuracy of training and verification in the database containing Gaussian Noise is less than the detection accuracy of the database without Gaussian Noise. Adjusting Gaussian Noise on the mean and the variance has different values from 0.01 to 0.09. The results indicate that greater Noise has a higher false classification, as the classification of thermal images reached 90% of false classification. Furthermore, during training, Deep Convolutional Neural Network Inception MV4 due to combining of actual thermal images with Noise thermal images with mean and variance values of 0.05 and 0.02 respectively and storing new thermal images in a single database. The results indicate that change in the accuracy of false detection, which is a very small mean value of 0.01 despite the change in variance with high values as shown in Figure 5. As for the second stage, the database of thermal images was used to which only Gaussian Noise was added without adding actual thermal images but storing modern thermal images as a new database. Therefore, the results show that the database has a slight impact on the accuracy of detection, as shown in Figure 6. Through the graph, they show a significant change in the mean of 0.01 and the variance of 0.01 compared to other values that maintained an accuracy of 100%. In the third stage, the MV4 Deep Convolutional Neural Network was verified on the main database consisting of 1800 thermal images without adding the Gaussian Noise. The results indicate a detection accuracy of 100%. Table 3 shows comparison between three thermal image databases after training Deep Convolutional Neural Network Inception MV4 after setting learning rate to 0.0001, training 10 epochs and using SGDM optimization method. Whereas best results indicated by table indicate that database does not contain Gaussian Noise is better than database that contains Gaussian Noise. Furthermore, average accuracy for database containing Gaussian Noise with actual thermal images and database containing only Gaussian Noise and the database without Gaussian Noise were 99.964%, 98.8% and 99.974% respectively. Table 4 shows addition of salt and pepper noise while adjusting variables from 0.1 to 0.3, with measuring accuracy several times and calculating average accuracy. The results show that deep convolutional neural network MV4 model was affected in accuracy average between 0.2 and 0.3 from 100% to 51.58%, without adding a filter in the pre-processing. Table 5 shows Variance of multiplicative noise, with variables adjusted from 0.2 to 0.8. With increase in noise, the classification accuracy was affected only in values 0.7, 0.8, which were 89% and 43%, respectively. On the other hand, when using speckle noise while adjusting variables from 0.2 to 0.8, the results show that classification accuracy decreased in value 0.8, however; it maintained its performance in remaining values as shown in Table 6. Table 7 indicates evaluation parameters such as accuracy, sensitivity, Specificity …etc. In conclusion, it is clear from the four tables that salt noise is the most influential on the classification accuracy of deep convolutional neural network MV4 without using filters during pre-processing. As for the use of filters in the pre-processing, it is clear that classification accuracy reached 97.999%. Table 8 focuses on five different values of Poisson noise setting (0.2 to 0.8) and examines the performance at various SNR (13.98 dB to 1.94 dB). The objective is to analyse how changes in SNR and affect the accuracy of the thermal image’s classification. Table 5 presents the average detection accuracy for various values in the Variance of multiplicative noise. In addition, it displays accuracies corresponding to different settings ranging from 0.2 to 0.8. On the other hand, in setting 0.2, the accuracy for all five measurements is a perfect 100%. Also, as variance increases to 0.3 and 0.4, the accuracy remains consistently high at 100%. However, when the variance reaches 0.5 and 0.6, we observe a slight decrease in accuracy, dropping to 99.9958% and 99.9026% respectively. Subsequently, in settings 0.7 and 0.8, accuracy notably decreases further to 89.6558% and 43.86% respectively. Finally, results indicate a trend where higher values of variance in multiplicative noise negatively impact the detection accuracy, with a significant drop observed beyond a variance of 0.6. In table 6 showing average detection accuracy of different values in speckle noise, various settings were tested with noise levels ranging from 0.2 to 0.8. The result indicates that at lower noise levels such as 0.2 and 0.3, the accuracy of detection was consistently at 100%. Moreover, as noise level increased, there were slight decreases in accuracy with most significant drop occurring at a noise level of 0.8. An accuracy 1 maintained a high accuracy rate across all noise levels, with a slight decrease to 95.4% at a noise level of 0.7. Accuracy 2 also showed high accuracy rates but experienced a significant drop to 44.5% at a noise level of 0.8. Accuracy 3 and Accuracy 4 had fluctuations in accuracy, with Accuracy 4 showing a noticeable decrease to 69% at a noise level of 0.6. Overall, the average accuracy across all settings decreased as the noise level increased, with the highest average accuracy of 100% observed at lower noise levels of 0.2 and 0.3. Also, average accuracy dropped to 62.114% at a noise level of 0.8, indicating a significant impact of higher noise levels on detection accuracy. Finaly, these results suggest that while the detection accuracy remains high at lower noise levels, there is a notable decline in performance as the noise level increases, highlighting the importance of noise reduction techniques in improving detection accuracy in speckle noise environments. The experiment was conducted by measuring the accuracy of the deep learning inception mv4 at different values of Poisson noise. The accuracy measurements were recorded for each SNR level, and the average accuracy was calculated across all measurements. The settings were denoted by SRN (Signal to Noise Ratio). The obtained data is presented in Table 6. From Table 6, at accuracy 1, it can be observed that at Poisson noise of 0.2, the accuracy is recorded as 100%, indicating perfect performance. However, as the Poisson noise increases, the accuracy decreases. At Poisson noise of 0.3 in Figure 7, the accuracy drops to 99.9%, suggesting a slight degradation in performance. As the Poisson noise increases further, the accuracy gradually improves, reaching 100% at Poisson noise of 0.6. Accuracy 2, Accuracy 3, Accuracy 4, Accuracy 5: These accuracy measurements follow a similar pattern to Accuracy 1. At lower Poisson noise values (0.2 to 0.4), the accuracy is considerably lower, indicating reduced performance. However, as the Poisson noise increases, the accuracy improves and eventually reaches 100% at a Poisson noise of 0.6. Moreover, the average accuracy across all measurements shows a consistent pattern. It is high (99.9% or 100%) at Poisson noise of 0.4 and above, indicating excellent overall performance. Finally, when the Poisson noise numbers used in the experiment are even decimal numbers, the corresponding results consistently exhibit a percentage of 100%. In contrast, when the noise numbers are odd decimals, there is a slight decrease in the observed percentage, specifically by 0.01%. Table 9 showcases the confusion matrices for three different deep learning models: Inception v3, Inception v4, and Inception mv4, each evaluated for their performance in breast cancer detection. Also, these matrices encapsulate the fundamental metrics crucial for assessing the models' classification accuracy. 4. Conclusions The study was conducted on a database for classifying thermal images with added Gaussian Noise with different values to evaluate the influence of added noise on the performance of the newly introduced Deep Convolutional Neural Network Inception MV4 in the classification of thermal images. It is evident that the thermal image database must be free from any noise that may affect the accuracy of classification. Salt noise was found to be having the most influence on the classification accuracy of the deep convolutional neural network MV4 without using filters during the pre-processing stage. However, adding noise filters are important in the preprocessing of thermal images before being fed into Deep Convolutional Neural Network Inception MV4 as they mitigate the noise effect to a large degree. Moreover, the accuracy of classifying noise-free thermal images reached 99.974%, which is the highest accuracy compared to the rest of the noisy databases. Future work calls for adding other types of noise to impact the image quality often experienced by stored and transferred thermal images to determine the best types of filters to process the noise before inserting the images into the Deep Convolutional Neural Network. Moreover, proposed future research directions include the development of representative datasets, the integration of segmented images into the training process, and the design of a lightweight convolutional neural network (CNN) model aimed at enhancing CNN performance for application across various disease-related topics. Declarations • Ethics approval and consent to participate : Not applicable • Consent for publication : Not applicable • Availability of data and materials : The datasets analysed during the current study are available in the DMR IR database repository, http://visual.ic.uff.br/dmi/ • Competing interests : authors declare no competing interests • Funding : The authors extend their appreciation to the Deanship of Scientific Research at King Khalid University for funding this work through the Research Group Program under Grant Number (R.G.P.2/400/44). • Authors' contributions : MAS contributed to the paper by conducting the literature review, writing the manuscript, revising it, creating figures and tables, interpreting the results, and providing critical feedback and review. MHH contributed to the paper by conducting the literature review, writing the manuscript, revising it, creating figures and tables, interpreting the results, and providing critical feedback and review. EAA contributed to the paper by conducting the literature review, writing the manuscript, revising it, creating figures and tables, interpreting the results, and providing critical feedback and review. MdR contributed to the paper by conducting the literature review, writing the manuscript, revising it, creating figures and tables, interpreting the results, and providing critical feedback and review. F. M. S contributed to the paper by conducting the literature review, writing the manuscript, revising it, creating figures and tables, interpreting the results, and providing critical feedback and review. YNS contributed to the paper by conducting the literature review, writing the manuscript, revising it, creating figures and tables, interpreting the results, and providing critical feedback and review. References Hanf V, Kreienberg R. Corpus uteri. 2020. 10.1007/978-3-662-11496-4_24 . Bini SA, Intelligence A, Learning M. Deep Learning, and Cognitive Computing: What Do These Terms Mean and How Will They Impact Health Care? J Arthroplasty. 2018;33(8):2358–61. 10.1016/j.arth.2018.02.067 . Yadav P, Jethani V. Breast thermograms analysisfor cancer detection using feature extraction and data mining technique, ACM Int. Conf. Proceeding Ser., 2016, 10.1145/2979779.2979866 . ud din NM, Dar RA, Rasool M, Assad A. Breast cancer detection using deep learning: Datasets, methods, and challenges ahead. Comput Biol Med. 2022;149:106073. 10.1016/j.compbiomed.2022.106073 . Salvi S, Kadam A. Breast Cancer Detection Using Deep learning and IoT Technologies. J Phys Conf Ser. 2021;1831(1). 10.1088/1742-6596/1831/1/012030 . Zhao P, Yoo I, Lancey R, Varghese E. Mobile applications for pain management: An app analysis for clinical usage. BMC Med Inf Decis Mak. 2019;19(1):1–10. 10.1186/s12911-019-0827-7 . Al Husaini MAS, Habaebi MH, Hameed SA, Islam MR, Gunawan TS. A Systematic Review of Breast Cancer Detection Using Thermography and Neural Networks. IEEE Access. 2020;8:208922–37. 10.1109/ACCESS.2020.3038817 . Al Husaini MAS, Habaebi MH, Islam MR, Gunawan TS. Self-detection of early breast cancer application with infrared camera and deep learning. Electron. 2021;10(20). 10.3390/electronics10202538 . Hiremath S, Karibasappa K. Neural Network Based Noise Identification in Digital Images. ACEEE Int J Netw Secur. 2011;02(03):3–6. Salami AM, Salih DM, Fadhil AF. Thermal Image Features and Noise Effects Analysis, in Proceedings of the 7th International Engineering Conference Research and Innovation Amid Global Pandemic, IEC 2021, Institute of Electrical and Electronics Engineers Inc., Feb. 2021, pp. 43–47. 10.1109/IEC52205.2021.9476100 . Liu Q, Liu Z, Yong S, Jia K, Razmjooy N. Computer-aided breast cancer diagnosis based on image segmentation and interval analysis. Automatika. 2020;61(3):496–506. 10.1080/00051144.2020.1785784 . Priyadharsini MS, HIGH DENSITY NOISE FILTER METHOD FOR DENOISING MAMMOGRAM BREAST,. Data cquisition Process no November. 2023. 10.5281/zenodo.776699 . Mulaveesala R, Dua G. Non-invasive and non-ionizing depth resolved infra-red imaging for detection and evaluation of breast cancer: a … Biomed. Phys Eng Express. 2018;2(5):1–5. 10.1088/2057-1976/2/5/055004 . Yousefi B, SPAER, et al. Sparse deep convolutional autoencoder model to extract low dimensional imaging biomarkers for early detection of breast cancer using dynamic thermography. Appl Sci. 2021;11(7). 10.3390/app11073248 . Ekici S, Jawzal H. Breast cancer diagnosis using thermography and convolutional neural networks, Med. Hypotheses, vol. 137, no. December 2019, p. 109542, 2020, 10.1016/j.mehy.2019.109542 . Kermani S, Samadzadehaghdam N, EtehadTavakol M, Optik, Stuttg.)., 126, 21, pp. 3288–94, 2015, 10.1016/j.ijleo.2015.08.007 . Dalmia A, Kakileti ST, Manjunath G. Exploring Deep Learning Networks for Tumour Segmentation in Infrared Images 2, 14 th Quant. InfraRed Thermogr. Conf., pp. 1–10, 2019, 10.1080/17686733.2019.1619355 . Roslidar R, et al. BreaCNet: A high-accuracy breast thermogram classifier based on mobile convolutional neural network. Math Biosci Eng. 2022;19(2):1304–31. 10.3934/mbe.2022060 . Gomathi P, Muniraj C, Periasamy PS. Digital infrared thermal imaging system based breast cancer diagnosis using 4D U-Net segmentation. Biomed Signal Process Control. 2023;85:104792. 10.1016/j.bspc.2023.104792 . Al Husaini MAS, Habaebi MH, Gunawan TS, Islam MR, Elsheikh EAA, Suliman FM. Thermal-based early breast cancer detection using inception V3, inception V4 and modified inception MV4. Neural Comput Appl. 2022;34(1):333–48. 10.1007/s00521-021-06372-1 . Szegedy C, Ioffe S, Vanhoucke V, Alemi A. Inception-v4, Inception-ResNet and the Impact of Residual Connections on Learning, Feb. 2016, [Online]. Available: http://arxiv.org/abs/1602.07261 . Szegedy C, Ioffe S, Vanhoucke V, Alemi AA. Inception-v4, Inception-ResNet and the Impact of Residual Connections on Learning, in Proceedings of the Thirty-First AAAI Conference on Artificial Intelligence (AAAI-17) Inception-v4, 2017, pp. 4278–4284. [Online]. Available: www.aaai.org. Goodman JW. Some fundamental properties of speckle*. J Opt Soc Am. 1976;66(11):1145. 10.1364/josa.66.001145 . Hiremath PS, P. T., and, Badiger S. Speckle Noise Reduction in Medical Ultrasound Images. Adv Break Ultrasound Imaging. 2013. 10.5772/56519 . Sudha S, Suresh GR, Sukanesh R. Speckle Noise Reduction in Ultrasound Images by Wavelet Thresholding based on Weighted Variance. Int J Comput Theory Eng no April. 2009;7–12. 10.7763/ijcte.2009.v1.2 . Buades A, Coll B, Morel JM. A non-local algorithm for image denoising, Proc. – 2005 IEEE Comput. Soc. Conf. Comput. Vis. Pattern Recognition, CVPR 2005, vol. II, no. July, pp. 60–65, 2005, 10.1109/CVPR.2005.38 . Rohit V, Ali J, A Comparative Study of Various Types of Image Noise and Efficient Noise Removal Techniques. Int J Adv Res Comput Sci Softw Eng. 2013;3(10):2277–128. Salmon J, Harmany Z, Deledalle CA, Willett R. Poisson noise reduction with non-local PCA. J Math Imaging Vis. 2014;48(2):279–94. 10.1007/s10851-013-0435-6 . Tables Table 1: Comparison between different Noise type and Deep learning models for early breast cancer detection. Source Noise type Deep learning model Accuracy [8] white Gaussian noise NA NA [9] Gaussian noise (3–20% noise) sparse deep convolutional autoencoder (SPAER) model 78.2% [10] salt and noise CNN 98.95% [11] Poisson noise NA NA [12] Gaussian noise cascaded CNN architecture 92% [13] Normal noise BreaCNet 100% [14] speckle noise Unet improvements with a 39.01% NA= Not Available Table 2: Comparison of parameters in Figure 4 such as Mean Noise Value, Standard Deviation of Noise and Noise Value in dB. Figure 4 Mean Noise Value Standard Deviation of Noise Noise Value in dB a 109.318 112.828 18.765 b 92.156 119.703 11.926 c 86.018 117.159 12.951 Table 3: Average detection accuracy of different databases. Epoch Gaussian Noise 0.05&0.02 DMRA DMRA+NOISE 1 96.06 100 99.91 2 96.06 100 99.91 3 96.96 100 100 4 99.82 99.82 100 5 100 100 99.82 6 99.82 99.92 100 7 99.64 100 100 8 99.82 100 100 9 100 100 100 10 99.82 100 100 Average 98.8 99.974 99.964 Table 4: Average detection accuracy of different value in salt and pepper noise Setting 0.1 0.2 0.26 0.27 0.28 0.29 0.3 accuracy 1 100 100 99.96 85 62 64 18 accuracy 2 100 100 99.99 99.9 74 99 99.9 accuracy 3 100 100 89.6 91.5 99.9 89 28 accuracy 4 100 100 99.92 98 99.4 75 52 accuracy 5 100 100 99.93 99.98 20 81 60 average accuracy 100 100 97.88 94.876 71.06 81.6 51.58 Table 5: Average detection accuracy of different value in Variance of multiplicative noise Setting 0.2 0.3 0.4 0.5 0.6 0.7 0.8 accuracy 1 100 100 100 100 99.996 99.8 0.6 accuracy 2 100 100 100 99.98 100 99.11 94 accuracy 3 100 100 100 99.999 99.999 99.999 30 accuracy 4 100 100 100 100 99.998 49.6 8 accuracy 5 100 100 100 100 99.52 99.77 86.7 average accuracy 100 100 100 99.9958 99.9026 89.6558 43.86 Table 6: Average detection accuracy of different value in speckle noise Setting 0.2 0.3 0.4 0.5 0.6 0.7 0.8 accuracy 1 100 100 100 100 99.998 95.4 97.77 accuracy 2 100 100 100 100 99.994 99.55 44.5 accuracy 3 100 100 100 100 99.999 99.84 3.3 accuracy 4 100 100 100 99.99 98.8 99.99 69 accuracy 5 100 100 100 100 99.968 99.92 96 average accuracy 100 100 100 99.998 99.7518 98.94 62.114 Table 7: Evaluation parameters for different type of database Type of database Accu Sens Spec P NPV FPR FNR LRN AUC EER F1 Gaussian Noise 0.05&0.02 96.06 0.970 0.952 0.946 0.974 0.048 0.030 0.031 0.961 0.039 0.958 DMRA+NOISE 99.82 0.996 1.000 1.000 0.997 0.000 0.004 0.004 0.998 0.002 0.998 DRMA 100 1.000 1.000 1.000 1.000 0.000 0.000 0.000 1.000 0.000 1.000 Table 8: Average detection accuracy of different value Poisson noise Poisson noise Setting 0.2 SRN 0.3 SRN 0.4 SRN 0.5 SRN 0.6 SRN 0.7 SRN 0.8 SRN accuracy 1 % 100 13.98 dB 99.9 10.46 dB 100 7.96 dB 99.9 6.02 dB 100 4.44 dB 99.9 3.10 dB 99.9 1.94 dB accuracy 2 % 100 99.9 100 99.9 100 99.9 99.9 accuracy 3 % 100 99.9 100 99.9 100 99.9 99.9 accuracy 4 % 100 99.9 100 99.9 100 99.9 99.9 accuracy 5 % 100 99.9 100 99.9 100 99.9 99.9 average accuracy % 100 99.9 100 99.9 100 99.9 99.9 Table 9: Confusion matrix for Inception v3, Inception v4 and Inception mv4 Deep learning Models Inception v3 Inception v4 Inception mv4 True Positive (TP) 255.000 171.000 171.000 True Negative (TN) 289.000 203.000 203.000 False Positive (FP) 3.000 0.000 0.000 False Negative (FN) 15.000 1.000 1.000 Additional Declarations No competing interests reported. Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-3978583","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":277556202,"identity":"50d70882-abad-463d-8cd0-adcb644f3a5e","order_by":0,"name":"Mohammed Abdulla Al Husaini","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAsElEQVRIiWNgGAWjYFACHgYJhgoIU4IELWdI1sLYRooW3fbegzd+zquT3XCA+eBtHgabaIJazM6cS7bs3XbYeMMBtmRrHoa03AaCWm7kmEnwbjuQuOEAj5k0D8Nh4rRI/p1TB9TC/414LdK8DcwgW9iI1HLmjLG1zLHDxjMPsxlbzjEgxi/Hewxvvqmpk+073vzwxpsKG8JaYICxgRlEGRCrHqyFBMWjYBSMglEwwgAAZig9F+TqiMgAAAAASUVORK5CYII=","orcid":"","institution":"International Islamic University Malaysia (IIUM)","correspondingAuthor":true,"submittingAuthor":false,"prefix":"","firstName":"Mohammed","middleName":"Abdulla Al","lastName":"Husaini","suffix":""},{"id":277556203,"identity":"3696764c-6e21-42ce-945e-4ab5cc07b9c5","order_by":1,"name":"Mohamed Hadi Habaebi","email":"","orcid":"","institution":"International Islamic University Malaysia (IIUM)","correspondingAuthor":false,"submittingAuthor":false,"prefix":"","firstName":"Mohamed","middleName":"Hadi","lastName":"Habaebi","suffix":""},{"id":277556204,"identity":"8907ad3d-5815-4948-8426-d22c2ff33603","order_by":2,"name":"Elfatih A.A. Elsheikh","email":"","orcid":"","institution":"King Khalid University (KKU)","correspondingAuthor":false,"submittingAuthor":false,"prefix":"","firstName":"Elfatih","middleName":"A.A.","lastName":"Elsheikh","suffix":""},{"id":277556205,"identity":"45991535-27d0-4853-b6ce-8cde892019fd","order_by":3,"name":"Md Rafiqul Islam","email":"","orcid":"","institution":"International Islamic University Malaysia (IIUM)","correspondingAuthor":false,"submittingAuthor":false,"prefix":"","firstName":"Md","middleName":"Rafiqul","lastName":"Islam","suffix":""},{"id":277556206,"identity":"264fad86-cdf7-4ea7-89b3-8cf61a7c6e97","order_by":4,"name":"F. M. Suliman","email":"","orcid":"","institution":"King Khalid University (KKU)","correspondingAuthor":false,"submittingAuthor":false,"prefix":"","firstName":"F.","middleName":"M.","lastName":"Suliman","suffix":""},{"id":277556207,"identity":"716b032b-ff0f-4293-b868-250927c3a0dd","order_by":5,"name":"Yousuf Nasser AL Husaini","email":"","orcid":"","institution":"Arab Open University","correspondingAuthor":false,"submittingAuthor":false,"prefix":"","firstName":"Yousuf","middleName":"Nasser AL","lastName":"Husaini","suffix":""}],"badges":[],"createdAt":"2024-02-22 12:32:33","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-3978583/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-3978583/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":52450908,"identity":"7e8d3b37-267d-4b42-9374-24d64ddb3fd3","added_by":"auto","created_at":"2024-03-11 19:10:15","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":53148,"visible":true,"origin":"","legend":"\u003cp\u003eInception MV4\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-3978583/v1/0ab0fcce1ba53c758d6e9c35.png"},{"id":52450906,"identity":"51a8faf0-bca4-4f1b-97e2-6c021309ef12","added_by":"auto","created_at":"2024-03-11 19:10:15","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":238768,"visible":true,"origin":"","legend":"\u003cp\u003eTraining Inception MV4\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-3978583/v1/ae5868026a8425d5d9c29000.png"},{"id":52450911,"identity":"1899b1ab-dfac-435c-bdb8-4aa7294e52e8","added_by":"auto","created_at":"2024-03-11 19:10:15","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":1285444,"visible":true,"origin":"","legend":"\u003cp\u003eThermal images unhealthy with different values in mean and variance\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-3978583/v1/20dd235a0cc1bd4c19fad553.png"},{"id":52450913,"identity":"48cf97c0-6754-4c35-8eca-1b581c9597e2","added_by":"auto","created_at":"2024-03-11 19:10:16","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":1116977,"visible":true,"origin":"","legend":"\u003cp\u003eHealthy thermal images: a) adding Gaussian Noise classified as Cancer, b) Denoising thermal image, c) classified thermal images after removing the noise as healthy 100%.\u003c/p\u003e","description":"","filename":"4.png","url":"https://assets-eu.researchsquare.com/files/rs-3978583/v1/5db9eed4dd60fa63fa798ec6.png"},{"id":52450907,"identity":"6f029a0e-2043-499e-81f0-51b34d2c5e8e","added_by":"auto","created_at":"2024-03-11 19:10:15","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":91031,"visible":true,"origin":"","legend":"\u003cp\u003eFalse detection accuracy with setting different values of Gaussian Noise\u003c/p\u003e","description":"","filename":"5.png","url":"https://assets-eu.researchsquare.com/files/rs-3978583/v1/31fea97939f6e32ee46cf140.png"},{"id":52450909,"identity":"1003c560-db84-4a62-b9ad-7d5d26650f06","added_by":"auto","created_at":"2024-03-11 19:10:15","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":41758,"visible":true,"origin":"","legend":"\u003cp\u003eDetection accuracy by using Gaussian noise database only.\u003c/p\u003e","description":"","filename":"6.png","url":"https://assets-eu.researchsquare.com/files/rs-3978583/v1/ee3050e1b531ec7ac6e3518d.png"},{"id":52450912,"identity":"77b0ae56-15d1-47a1-a6cd-c9bbd2236d77","added_by":"auto","created_at":"2024-03-11 19:10:15","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":1181682,"visible":true,"origin":"","legend":"\u003cp\u003eThermal images a) original thermal image b) thermal image after adding Poisson noise with value 0.3\u003c/p\u003e","description":"","filename":"7.png","url":"https://assets-eu.researchsquare.com/files/rs-3978583/v1/19d034b6b2ab05cba4ad3dd8.png"},{"id":63519614,"identity":"7e69c4d4-70ef-4296-a439-ce857a4fcff9","added_by":"auto","created_at":"2024-08-29 05:40:02","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":4301437,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-3978583/v1/d3b7d5ce-9ffe-41f9-be3e-fd84d943e184.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Evaluating the Effect of Noisy Thermal Images On the Detection of Early Breast Cancer Using Deep Learning","fulltext":[{"header":"1. Introduction","content":"\u003cp\u003eAccording to the World Health Organization, statistics indicate an increase in the incidence of breast cancer. The statistics mentioned in 2020 indicate that there are 2.088\u0026nbsp;million new cases of breast cancer, and the most important reason is late detection [\u003cspan class=\"CitationRef\"\u003e1\u003c/span\u003e]. Machine learning and deep learning have emerged as transformative technologies in the field of medical sciences, revolutionizing various aspects of healthcare, including disease diagnosis in early stages and treatment [\u003cspan class=\"CitationRef\"\u003e2\u003c/span\u003e]. In addition, these advanced techniques have played a pivotal role in improving accuracy and efficiency in the realm of breast cancer detection [\u003cspan class=\"CitationRef\"\u003e3\u003c/span\u003e][\u003cspan class=\"CitationRef\"\u003e4\u003c/span\u003e][\u003cspan class=\"CitationRef\"\u003e5\u003c/span\u003e]. Moreover, with vast amounts of accurate medical database available, machine learning algorithms can analyse intricate patterns and anomalies within mammograms, ultrasound images, and pathology slides, aiding in the early detection of breast cancer. Deep learning models have demonstrated remarkable capabilities in image recognition and classification tasks, enabling healthcare professionals to identify suspicious lesions with high precision. Moreover, these technologies have the potential to assist radiologists and pathologists by providing them with decision support tools, thereby enhancing diagnostic accuracy and reducing the likelihood of human error [\u003cspan class=\"CitationRef\"\u003e6\u003c/span\u003e]. Recently, thermal imaging technology has been utilized to detect breast cancer using deep learning. Therefore, it is a safe, harmless, and promising technique for early detection [\u003cspan class=\"CitationRef\"\u003e7\u003c/span\u003e] This technology has become a home diagnostic tool that enables patients to track tumor status periodically [\u003cspan class=\"CitationRef\"\u003e8\u003c/span\u003e]. Therefore, it is subject to Noise in some cases that may not comply with thermal imaging procedures. Therefore, one of the factors affecting the classification of thermal images in deep learning is location and size of tumor in breast and the presence of Noise in thermal images. This study will explain the im-pact of Gaussian Noise on thermal images and on databases. Image noise, which is undesired information within an image, can manifest at various stages like image capture, transmission, or processing. In addition, a comprehensive understanding of the noise\u0026apos;s characteristics is essential to effectively eliminate noise from a noisy image. On the other hand, attempts at noise removal may lead to image blurring [\u003cspan class=\"CitationRef\"\u003e9\u003c/span\u003e]. The study in [\u003cspan class=\"CitationRef\"\u003e10\u003c/span\u003e] explores the diverse effects of noise types, such as Gaussian, Salt \u0026amp; Pepper, Speckle, and Poison, on thermal image features\u0026mdash;specifically Histogram, Texture, and Trans-form-based features. Evaluating both pre- and post-noise application, Signal-to-Noise Ratio and Mean Absolute Error serve as quality criteria and Results underscore the substantial influence of the Mode feature in thermal images compared to regular im-ages.\u003c/p\u003e\n\u003cp\u003eThe fundamental idea is to enhance the conventional Laplacian of Gaussian filter by incorporating interval analysis to account for intensity uncertainties. Also, experimental evaluations were conducted on MIAS, a widely used breast cancer database, for medical image segmentation. Finally, system\u0026apos;s performance was assessed in comparison to Prewitt, LoG, and Canny filters using the Peak Signal-to-Noise Ratio (PSNR) [\u003cspan class=\"CitationRef\"\u003e11\u003c/span\u003e]. On the other hand, images captured from a Mammogram might exhibit noise caused by variations in lighting and sensor inaccuracies. Cancer\u0026rsquo;s noise can be effectively removed without compromising the image\u0026apos;s boundaries and small details, accurate diagnoses of breast cancer can be facilitated through imaging technology [\u003cspan class=\"CitationRef\"\u003e12\u003c/span\u003e].\u003c/p\u003e\n\u003cdiv id=\"Sec2\" class=\"Section2\"\u003e\n \u003ch2\u003e1.1 Related works\u003c/h2\u003e\n \u003cp\u003eThe study [\u003cspan class=\"CitationRef\"\u003e13\u003c/span\u003e] referred to the use of the numerical simulation system for the location of tumors in the breast on different sites and the addition of thermal cooling. Gaussian Noises were added to thermal images. In addition, only four tumor sites were placed in the breast at different positions. The obtained results claim that the time-domain phase approach shows improved detection capabilities over those of raw heat. The study conducted by the researcher [\u003cspan class=\"CitationRef\"\u003e14\u003c/span\u003e] proposed a new model of Deep Convolutional Neural Network called SPARs. This model has been proposed for low-dimensional deep heat extraction. In addition, 208 clinical breast cancer thermal images were used with the DMR IR database. Moreover, Gaussian Noise was added at a rate of 3 to 20% on the thermal images, but after encoding the results showed a slight decrease in the accuracy of the thermal images. SPAER demonstrated great robustness when tested for additive Gaussian Noise conditions (3\u0026ndash;20% Noise), as assessed by its signal-to-noise ratio (SNR). The results indicate a high performance of SPAER to maintain thermal heterogeneity, and it can be used as an in vivo non-invasive tool that aids CBE in the early detection of breast cancer.\u003c/p\u003e\n \u003cp\u003eIn [\u003cspan class=\"CitationRef\"\u003e15\u003c/span\u003e] used the Deep Convolutional Neural Network and BIOS optimization algorithm. In addition, thermal images used were 3895, divided as follows: 3098 healthy and 757 cancers. also referred to the processing of images with added Noise. The results indicate an increase in the accuracy rate from 97.91 to 98.95%. The researcher [\u003cspan class=\"CitationRef\"\u003e16\u003c/span\u003e] refers to processing thermal images and the removal of Noise by using the soft wavelet threshold before color segmentation. In addition, the proposed method was applied to 50 thermal images after applying pre-treatment. Clinical thermal images were used by a 640 x 480 pixels thermal camera. The results indicate that carefully designed color maps visually improve the thermal image, and improve pest detection and interpretation without using any algorithms or deep learning. The study presented by [\u003cspan class=\"CitationRef\"\u003e17\u003c/span\u003e] utilized Cascade Deep Convolutional Neural Network to classify thermal images of breast cancer. The database consists of 900 thermal images. Gaussian Noise has been added to increase the database of thermal images. In addition, Deep Convolutional Neural Network was trained on thermal images for automatic classification. CNN was trained by setting 50 epochs and learning rate of 0.5e-3. The results indicate that accuracy reached 92%.\u003c/p\u003e\n \u003cp\u003eThe researchers [\u003cspan class=\"CitationRef\"\u003e18\u003c/span\u003e] referred to utilize of the deep convolutional neural network BreaCNet model, which is a modified deep convolutional neural network Shuffle Net model. Also, it can extract around 6\u0026nbsp;million features. In addition, 1302 thermal images were used for training 90% and testing 10% which were collected from DMR IR database. Moreover, Sobel kernel was used to filter the edges in the region of interest (ROI) and convert color thermal images into grayscale. In addition, the modified model deep convolutional neural network was tuned learning 1e -3 with 75 epoch and SGDM optimization method was used. Results indicated the accuracy of the detection reached 100% compared to Mobilnet, which reached 98%. Furthermore, a Gaussian filter has been added and smartphone app has been created with size 22MB.\u003c/p\u003e\n \u003cp\u003eIn [\u003cspan class=\"CitationRef\"\u003e19\u003c/span\u003e] a new breast cancer detection method, DBC-4D U-Net-DITI, utilizing digital infrared thermal imaging. Also, it employs Altered Phase Preserving Dynamic Range Compression (APPDRC) for preprocessing, optimizes 4D U-Net weights using Glow-worm Swarm Optimization Algorithm (GSOA) for segmentation, reduce speckle noise, and employs a Binarized Spiking Neural Network (BSNN) for pathology stage classification. In conclusion, results indicate in comparison to existing methods like DBC-CSSA-DITI, DBC-MPA-DITI, and DBC-CNN-DITI, the proposed approach demonstrates substantial improvements with a 39.01%, 28.34%, and 37.45% accuracy boost and a 17.12%, 24.12%, and 32.07% precision enhancement, respectively.\u003c/p\u003e\n \u003cp\u003eIt is clear from the previous studies and Table 1 that Gaussian Noise was used in these studies. However, these studies lack clarity of noise data. Therefore, the proposed approach introduces 3 different types of noise with different values to a database consisting of thermal images. In addition, the effect of an added set of noise values on the classification capabilities of Deep Convolutional Neural Network Inception MV4 is studied. In conclusion, filters were added to mitigate the effect of noise from thermal images before classification.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e\n \u003ch2\u003e1.2 Motivation\u003c/h2\u003e\n \u003cp\u003eBreast cancer is a significant health concern worldwide, and early detection con-tributes to improving patient outcomes. Thermal imaging, as non-invasive, shows promise in detecting breast cancer in early stages. However, thermal images are susceptible to different types of noises, such as Gaussian noise, speckle noise, salt and pepper noise, and Poisson noise which will affect image quality. Therefore, under-standing the effects of different types of noises on thermal images of breast cancer are essential for developing robust and reliable deep learning algorithms for accurate detection and diagnosis.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec4\" class=\"Section2\"\u003e\n \u003ch2\u003e1.3 Contribution\u003c/h2\u003e\n \u003cp\u003eThe evaluation of Gaussian noise effects in thermal images of breast cancer and its impact on deep learning methods have an important contribution:\u003c/p\u003e\n \u003cp\u003e(i) The study provides a thorough evaluation of the effects of different types of noises on thermal images specifically related to breast cancer. By studying the effect of noise on image quality, important features, and overall diagnostic performance, enhances our understanding of the challenges associated with noisy thermal images.\u003c/p\u003e\n \u003cp\u003e(ii) The research examines the performance of deep learning algorithms, such as convolutional neural networks (CNNs), in the presence of Gaussian noise. By quantifying the degradation in accuracy, sensitivity, and specificity caused by noise, it provides insights into the limitations of existing deep learning models and highlights the need for noise-aware techniques.\u003c/p\u003e\n \u003cp\u003e(iii) Based on the evaluation results, the study investigates and proposes effective strategies to mitigate the adverse effects of different types of noise on thermal images. This could include preprocessing techniques, noise reduction algorithms, or modified deep learning architectures that are robust to noise, ultimately improving the accuracy and reliability of breast cancer detection systems.\u003c/p\u003e\n \u003cp\u003e(iv) The findings and recommendations from this research have practical implications for medical professionals, researchers, and developers working on thermal imaging-based breast cancer diagnosis. The study can guide the development of noise-robust deep learning models, contribute to the design of optimized imaging protocols, and assist in the implementation of quality control measures for thermal imaging devices.\u003c/p\u003e\n \u003cp\u003eBy addressing the motivation and making these significant contributions, the evaluation of Gaussian noise effects in thermal images of breast cancer and deep learning advances our understanding of the challenges associated with noisy thermal images and helps pave the way for improved early detection and diagnosis of breast cancer.\u003c/p\u003e\n\u003c/div\u003e"},{"header":"2. Materials and Methods","content":"\u003cp\u003eIn this section, we will introduce the deep learning framework used in this paper. Namely, inception v3, inception v4, and modified inception v4 models [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e] are introduced below.\u003c/p\u003e \u003cdiv id=\"Sec6\" class=\"Section2\"\u003e \u003ch2\u003e2.1 Inception v3\u003c/h2\u003e \u003cp\u003eInception-v3 is a convolutional neural network (CNN) architecture that has been created in 2015 by Google researchers. It is part of the Inception family of CNN models and is designed for image recognition and classification tasks. Inception-v3 is characterized by its deep architecture, consisting of 48 layers in total. It uses a combination of different size of filters in convolutional layers, max pooling, average pooling layers, and fully connected layers. The architecture is based on the idea of \"inception modules,\" which are building blocks that allow for efficient multi-scale feature extraction. The main innovation of Inception-v3 lies in its inception modules. These modules use various filter sizes (1x1, 3x3, 5x5) to capture information at different spatial scales and process them in parallel. This helps the network capture both fine-grained details and high-level features. To reduce the computational complexity of the network, Incep-tion-v3 uses 1x1 convolutions as dimensionality reduction layers. These layers reduce the number of input channels before applying larger convolutional filters, thereby im-proving efficiency without significant loss of information. Inception-v3 includes auxiliary classifiers at intermediate layers. These classifiers help combat the vanishing gradient problem during training and provide regularization. They also serve as additional sources of gradients during backpropagation, aiding in the overall training process. Inception-v3 is typically pre-trained on large-scale datasets like ImageNet, which contains millions of labelled images from various categories. After pre-training, the model can be fine-tuned on smaller datasets for specific image recognition tasks, adapting its learned features to the new dataset. Inception-v3 achieves impressive performance on image classification benchmarks. In the ImageNet Large Scale Visual Recognition Challenge (ILSVRC) 2015, it achieved a top-5 error rate of 3.46%, surpassing its predecessor models and demonstrating state-of-the-art accuracy at the time. Inception-v3 and its subsequent versions have been widely adopted and used as a backbone architecture in many computer vision applications, including object recognition, image segmentation, and transfer learning tasks. Its design principles have also influenced the development of other CNN architectures, such as Inception-v4 and Inception-ResNet [\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e].\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec7\" class=\"Section2\"\u003e \u003ch2\u003e2.2 Inception v4\u003c/h2\u003e \u003cp\u003eInception-v4 is an advanced convolutional neural network architecture that im-proves upon previous versions of the Inception family. It simplifies the overall structure, introduces a stem layer, and incorporates a greater number of inception modules compared to Inception-v3. The architecture of Inception-v4 is characterized by a more uniform and simplified design. The network consists of a comprehensive planner and stem configuration, along with 4 inception A layers, 7 inception B layers, and 3 inception C layers [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e].\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003e2.3 Inception mv4\u003c/h2\u003e \u003cp\u003eAn examination was conducted to compare \"inception B\" in Inception V4 to the updated version in MV4. The modifications made were limited to \"inception B\" as shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e1\u003c/span\u003e. First, a convolutional layer was added beneath the average pooling layer. The number of filters was increased from 128 to 256 to maintain the number of features extracted. Then, two parallel convolutional layers were added under the layer with 192 filters, both with the same size and number of filters. After that, the remaining layers in inception B were removed to keep the number of extracted features. Finally, Inception B has 7 groups, each with a set of layers. All 7 groups were changed with the same adjustments mentioned before. For further details on the framework, readers are advised to consult [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e].\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec9\" class=\"Section2\"\u003e \u003ch2\u003e2.4 DMR IR Database for Mastology Research with Infrared Image\u003c/h2\u003e \u003cp\u003eThe patient must abstain from hot liquid, activity, and applying any creams to their breasts and underarms at least 2 hours before the exam. Moreover, the exam room should be kept at a temperature between 20\u0026ndash;22\u0026deg;C. Also, Patient Preparation: In the exam room, the patient should remove any jewellery or accessories that could affect the thermal image. Additionally, the camera should be positioned 1 meter from the patient. The acquisition should be dynamic with the patient facing the camera. IR images are taken using a FLIR SC620 with a sensitivity of less than 0.04\u0026deg;C and a temperature range of -40\u0026deg;C to 500\u0026deg;C. It can be accessed through an online interface (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://visual.ic.uff.br/dmi\u003c/span\u003e\u003cspan address=\"http://visual.ic.uff.br/dmi\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e) that allows for easy management and retrieval of information from breast exams and patient clinical data.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec10\" class=\"Section2\"\u003e \u003ch2\u003e2.5 Speckle noise model\u003c/h2\u003e \u003cp\u003eA speckle pattern that appears in polarized monochromatic light can be viewed as being caused by a classical random walk in the complex plane [\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e]. Speckle noise is a type of noise that appears in images as a rough and grainy pattern resembling salt and pepper noise. It has the greatest impact on thermal image features, while Salt \u0026amp; Pepper noise has the lower impact [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]. This noise referred to is a multiplicative noise with a granular form. The presence of speckle noise in images is unwanted as it degrades the thermal image quality by impacting the edges and local details between different organs, which are crucial for diagnostic purposes [\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e]. Speckle noise in the medical literature is known as \"texture\" and might hold valuable diagnostic data. The ideal level of speckle smoothing is largely influenced by the expert's expertise and the intended use. In automatic segmentation, preserving the crispness of boundaries between several image areas is commonly selected while reducing the speckled texture [\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e].\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003e2.6 Gaussian noise model\u003c/h2\u003e \u003cp\u003eGaussian noise, also known as additive white Gaussian noise (AWGN), is a common type of random noise that occurs in many imaging and signal processing applications. It is characterized by random variations in pixel intensities that follow a Gaussian distribution. It is often introduced during image acquisition, transmission, or processing due to various factors like electronic noise, quantization errors, or sensor imperfections. The noise appears as a random fluctuation around the true pixel value, resulting in a smooth distribution of noise values across the image. Moreover, the properties of Gaussian noise make it an important noise model in many theoretical and practical ap-plications. In addition, the Gaussian distribution is symmetric and bell-shaped, with most noise values concentrated around the mean value and fewer values occurring at the extremes. However, the noise values are statistically independent at each pixel, and their distribution is characterized by the mean and standard deviation parameters. The choice of denoising technique depends on the noise characteristics, the desired level of noise reduction, and the trade-off between noise removal and preservation of image details [\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e].\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003e2.7 Salt and pepper noise\u003c/h2\u003e \u003cp\u003eSalt and pepper noise is also known as impulse noise, and it is a common type of image noise that can affect the quality and clarity of digital thermal images. It is characterized by the occurrence of randomly scattered white and black pixels, resembling grains of salt and pepper. It can occur due to various factors such as transmission errors, faulty sensors, or electronic interference during the image acquisition or processing stages. It typically manifests as isolated bright white pixels (salt) and dark black pixels (pepper) randomly scattered throughout the image. The presence of salt and pepper noise can significantly affect image analysis and processing tasks, as it introduces un-wanted artifacts and distorts the overall appearance of the image. Therefore, it is crucial to employ denoising techniques to mitigate the effects of this noise and restore image quality. When selecting a denoising method, it is important to consider the trade-off between noise removal and preservation of important image features. Different methods may yield varying results depending on the characteristics of the noise and the desired image quality [\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e].\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003e2.8 Poisson noise\u003c/h2\u003e \u003cp\u003ePoisson noise is a type of random variation that commonly occurs in images captured under low-light conditions or when using imaging devices with photon-counting sensors, like medical imaging devices. Poisson noise is caused by the random nature of photon arrival at the sensor or detector. In addition, it manifests as a variation in the number of photons detected at each pixel, resulting in intensity fluctuations in the im-age. The Poisson distribution models the probability of observing a certain number of events (photons) in each interval of time or space. In digital images, Poisson noise appears as random variations in pixel intensities, with brighter areas having a higher probability of more photons and darker areas having a lower probability. This noise can degrade image quality by reducing contrast, introducing unwanted variations, and affecting image analysis and processing tasks. Moreover, it is inherently different from other types of noise, such as Gaussian noise, due to its statistical properties. Therefore, specialized denoising techniques are required to effectively address this specific type of noise [\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e].\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec14\" class=\"Section2\"\u003e \u003ch2\u003e2.9 Experiment setup\u003c/h2\u003e \u003cp\u003eThe experiments conducted to detect breast cancer at an early stage were by adding Gaussian Noises of various proportions. The results obtained were compared with the results in previous studies. In addition, the experiments were divided into four stages. The first stage is to use Gaussian Noise in the thermal images database only. The second stage is to use the original database without adding noise which is used to verify the modified Deep Convolutional Neural Network inception MV4. The third stage is the merging of original thermal images from the DMR IR database with thermal images with added Gaussian Noise. The fourth stage is to add Noise filters before classifying thermal images in a modified Deep Convolutional Neural Network inception MV4. As for the devices used in the experiments: a computer with high specifications, 32 GB RAM, Core i7, and a storage capacity of 1 TB. In addition, MATLAB version 2020a software and graphics processor unit. Moreover, thermal images from the DMR IR database were utilized, with a total of 1800 thermal images categorized as healthy (800 thermal images) and unhealthy (1000 thermal images). The deep Convolutional Neural Network inception MV4 model was used because it has high accuracy compared to other neural networks as shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e1\u003c/span\u003e [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e].\u003c/p\u003e \u003c/div\u003e"},{"header":"3. Results and Discussion","content":"\u003cp\u003eFigure 1 indicates that the change in Gaussian Noise values affects to classification quality of the Deep Convolutional Neural Network, as healthy thermal images were classified as diseased. Gaussian Noise was added at the following values 0.05 and 0.02 in all thermal images of the database and the new database was saved. In addition, Deep Convolutional Neural Network Inception MV4 was set to a learning rate of 0.0001, SGDM optimization method and training 10 epochs see Figure 2. As for the second stage, Deep Convolutional Neural Network was trained using the same settings, but without adding Gaussian Noise. The third stage was the training of Deep Convolutional Neural Network on the same settings, but by adding actual thermal images with added Gaussian Noise.\u003c/p\u003e\n\u003cp\u003eThe following graph shows the amount of change in Noise compared to the accuracy of classification in each of DMR IR and DMR IR databases with Noise and with Gaussian Noise. Where the results indicate that the database without Noise has best solution for classification. As for database with Noise in any of the thermal images, filters are added during image processing and before entering Deep Convolutional Neural Network. Figure 3 is clear from the comparison between thermal image 1, thermal image 2, and thermal image 3 constitutes the difference between the addition of Gaussian Noise and the actual thermal image. Moreover, it is clear from Figure 4 gradient in addition to Gaussian Noise and the extent of its impact on the accuracy of classification. Table 2 shows that the analysis reveals a mean noise value, a standard deviation of noise, and a noise value in dB from Figure 4. These parameters impact on the classification accuracy in thermal images. Fourth stage: thermal image processing filters built before being inserted into Deep Convolutional Neural Network. Detection accuracy was calculated and found to highlight the effect of filters on the classification using Deep Convolutional Neural Network. The results indicate that the accuracy of training and verification in the database containing Gaussian Noise is less than the detection accuracy of the database without Gaussian Noise. Adjusting Gaussian Noise on the mean and the variance has different values from 0.01 to 0.09. The results indicate that greater Noise has a higher false classification, as the classification of thermal images reached 90% of false classification. Furthermore, during training, Deep Convolutional Neural Network Inception MV4 due to combining of actual thermal images with Noise thermal images with mean and variance values of 0.05 and 0.02 respectively and storing new thermal images in a single database. The results indicate that change in the accuracy of false detection, which is a very small mean value of 0.01 despite the change in variance with high values as shown in Figure 5.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eAs for the second stage, the database of thermal images was used to which only Gaussian Noise was added without adding actual thermal images but storing modern thermal images as a new database. Therefore, the results show that the database has a slight impact on the accuracy of detection, as shown in Figure 6. Through the graph, they show a significant change in the mean of 0.01 and the variance of 0.01 compared to other values that maintained an accuracy of 100%. In the third stage, the MV4 Deep Convolutional Neural Network was verified on the main database consisting of 1800 thermal images without adding the Gaussian Noise. The results indicate a detection accuracy of 100%.\u003c/p\u003e\n\u003cp\u003eTable 3 shows comparison between three thermal image databases after training Deep Convolutional Neural Network Inception MV4 after setting learning rate to 0.0001, training 10 epochs and using SGDM optimization method. Whereas best results indicated by table indicate that database does not contain Gaussian Noise is better than database that contains Gaussian Noise. Furthermore, average accuracy for database containing Gaussian Noise with actual thermal images and database containing only Gaussian Noise and the database without Gaussian Noise were 99.964%, 98.8% and 99.974% respectively.\u003c/p\u003e\n\u003cp\u003eTable 4 shows addition of salt and pepper noise while adjusting variables from 0.1 to 0.3, with measuring accuracy several times and calculating average accuracy. The results show that deep convolutional neural network MV4 model was affected in accuracy average between 0.2 and 0.3 from 100% to 51.58%, without adding a filter in the pre-processing. Table 5 shows Variance of multiplicative noise, with variables adjusted from 0.2 to 0.8. With increase in noise, the classification accuracy was affected only in values 0.7, 0.8, which were 89% and 43%, respectively. On the other hand, when using speckle noise while adjusting variables from 0.2 to 0.8, the results show that classification accuracy decreased in value 0.8, however; it maintained its performance in remaining values as shown in Table 6. Table 7 indicates evaluation parameters such as accuracy, sensitivity, Specificity \u0026hellip;etc.\u003c/p\u003e\n\u003cp\u003eIn conclusion, it is clear from the four tables that salt noise is the most influential on the classification accuracy of deep convolutional neural network MV4 without using filters during pre-processing. As for the use of filters in the pre-processing, it is clear that classification accuracy reached 97.999%. Table 8 focuses on five different values of Poisson noise setting (0.2 to 0.8) and examines the performance at various SNR (13.98 dB to 1.94 dB). The objective is to analyse how changes in SNR and affect the accuracy of the thermal image\u0026rsquo;s classification.\u003c/p\u003e\n\u003cp\u003eTable 5 presents the average detection accuracy for various values in the Variance of multiplicative noise. In addition, it displays accuracies corresponding to different settings ranging from 0.2 to 0.8. On the other hand, in setting 0.2, the accuracy for all five measurements is a perfect 100%. Also, as variance increases to 0.3 and 0.4, the accuracy remains consistently high at 100%. However, when the variance reaches 0.5 and 0.6, we observe a slight decrease in accuracy, dropping to 99.9958% and 99.9026% respectively. Subsequently, in settings 0.7 and 0.8, accuracy notably decreases further to 89.6558% and 43.86% respectively. Finally, results indicate a trend where higher values of variance in multiplicative noise negatively impact the detection accuracy, with a significant drop observed beyond a variance of 0.6.\u003c/p\u003e\n\u003cp\u003eIn table 6 showing average detection accuracy of different values in speckle noise, various settings were tested with noise levels ranging from 0.2 to 0.8. The result indicates that at lower noise levels such as 0.2 and 0.3, the accuracy of detection was consistently at 100%. Moreover, as noise level increased, there were slight decreases in accuracy with most significant drop occurring at a noise level of 0.8. An accuracy 1 maintained a high accuracy rate across all noise levels, with a slight decrease to 95.4% at a noise level of 0.7. Accuracy 2 also showed high accuracy rates but experienced a significant drop to 44.5% at a noise level of 0.8. Accuracy 3 and Accuracy 4 had fluctuations in accuracy, with Accuracy 4 showing a noticeable decrease to 69% at a noise level of 0.6. Overall, the average accuracy across all settings decreased as the noise level increased, with the highest average accuracy of 100% observed at lower noise levels of 0.2 and 0.3. Also, average accuracy dropped to 62.114% at a noise level of 0.8, indicating a significant impact of higher noise levels on detection accuracy. Finaly, these results suggest that while the detection accuracy remains high at lower noise levels, there is a notable decline in performance as the noise level increases, highlighting the importance of noise reduction techniques in improving detection accuracy in speckle noise environments.\u003c/p\u003e\n\u003cp\u003eThe experiment was conducted by measuring the accuracy of the deep learning inception mv4 at different values of Poisson noise. The accuracy measurements were recorded for each SNR level, and the average accuracy was calculated across all measurements. The settings were denoted by SRN (Signal to Noise Ratio). The obtained data is presented in Table 6. From Table 6, at accuracy 1, it can be observed that at Poisson noise of 0.2, the accuracy is recorded as 100%, indicating perfect performance. However, as the Poisson noise increases, the accuracy decreases. At Poisson noise of 0.3 in Figure 7, the accuracy drops to 99.9%, suggesting a slight degradation in performance. As the Poisson noise increases further, the accuracy gradually improves, reaching 100% at Poisson noise of 0.6. Accuracy 2, Accuracy 3, Accuracy 4, Accuracy 5: These accuracy measurements follow a similar pattern to Accuracy 1. At lower Poisson noise values (0.2 to 0.4), the accuracy is considerably lower, indicating reduced performance. However, as the Poisson noise increases, the accuracy improves and eventually reaches 100% at a Poisson noise of 0.6. Moreover, the average accuracy across all measurements shows a consistent pattern. It is high (99.9% or 100%) at Poisson noise of 0.4 and above, indicating excellent overall performance. Finally, when the Poisson noise numbers used in the experiment are even decimal numbers, the corresponding results consistently exhibit a percentage of 100%. In contrast, when the noise numbers are odd decimals, there is a slight decrease in the observed percentage, specifically by 0.01%. Table 9 showcases the confusion matrices for three different deep learning models: Inception v3, Inception v4, and Inception mv4, each evaluated for their performance in breast cancer detection. Also, these matrices encapsulate the fundamental metrics crucial for assessing the models\u0026apos; classification accuracy.\u003c/p\u003e"},{"header":"4. Conclusions","content":"\u003cp\u003eThe study was conducted on a database for classifying thermal images with added Gaussian Noise with different values to evaluate the influence of added noise on the performance of the newly introduced Deep Convolutional Neural Network Inception MV4 in the classification of thermal images. It is evident that the thermal image database must be free from any noise that may affect the accuracy of classification. Salt noise was found to be having the most influence on the classification accuracy of the deep convolutional neural network MV4 without using filters during the pre-processing stage. However, adding noise filters are important in the preprocessing of thermal images before being fed into Deep Convolutional Neural Network Inception MV4 as they mitigate the noise effect to a large degree. Moreover, the accuracy of classifying noise-free thermal images reached 99.974%, which is the highest accuracy compared to the rest of the noisy databases. Future work calls for adding other types of noise to impact the image quality often experienced by stored and transferred thermal images to determine the best types of filters to process the noise before inserting the images into the Deep Convolutional Neural Network. Moreover, proposed future research directions include the development of representative datasets, the integration of segmented images into the training process, and the design of a lightweight convolutional neural network (CNN) model aimed at enhancing CNN performance for application across various disease-related topics.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u0026bull; \u003cstrong\u003eEthics approval and consent to participate\u003c/strong\u003e: Not applicable\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u0026bull; \u003cstrong\u003eConsent for publication\u003c/strong\u003e : Not applicable\u003c/p\u003e\n\u003cp\u003e\u0026bull; \u003cstrong\u003eAvailability of data and materials\u003c/strong\u003e : The datasets analysed during the current study are available in the DMR IR database repository, http://visual.ic.uff.br/dmi/\u003c/p\u003e\n\u003cp\u003e\u0026bull; \u003cstrong\u003eCompeting interests\u003c/strong\u003e: authors declare no competing interests\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u0026bull; \u003cstrong\u003eFunding\u003c/strong\u003e: The authors extend their appreciation to the Deanship of Scientific Research at King Khalid University for funding this work through the Research Group Program under Grant Number (R.G.P.2/400/44).\u003c/p\u003e\n\u003cp\u003e\u0026bull; \u003cstrong\u003eAuthors\u0026apos; contributions\u003c/strong\u003e:\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eMAS contributed to the paper by conducting the literature review, writing the manuscript, revising it, creating figures and tables, interpreting the results, and providing critical feedback and review.\u003c/p\u003e\n\u003cp\u003eMHH contributed to the paper by conducting the literature review, writing the manuscript, revising it, creating figures and tables, interpreting the results, and providing critical feedback and review.\u003c/p\u003e\n\u003cp\u003eEAA contributed to the paper by conducting the literature review, writing the manuscript, revising it, creating figures and tables, interpreting the results, and providing critical feedback and review.\u003c/p\u003e\n\u003cp\u003eMdR \u0026nbsp; contributed to the paper by conducting the literature review, writing the manuscript, revising it, creating figures and tables, interpreting the results, and providing critical feedback and review.\u003c/p\u003e\n\u003cp\u003eF. M. S contributed to the paper by conducting the literature review, writing the manuscript, revising it, creating figures and tables, interpreting the results, and providing critical feedback and review.\u003c/p\u003e\n\u003cp\u003eYNS contributed to the paper by conducting the literature review, writing the manuscript, revising it, creating figures and tables, interpreting the results, and providing critical feedback and review.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eHanf V, Kreienberg R. Corpus uteri. 2020. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/978-3-662-11496-4_24\u003c/span\u003e\u003cspan address=\"10.1007/978-3-662-11496-4_24\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBini SA, Intelligence A, Learning M. Deep Learning, and Cognitive Computing: What Do These Terms Mean and How Will They Impact Health Care? J Arthroplasty. 2018;33(8):2358\u0026ndash;61. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.arth.2018.02.067\u003c/span\u003e\u003cspan address=\"10.1016/j.arth.2018.02.067\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYadav P, Jethani V. Breast thermograms analysisfor cancer detection using feature extraction and data mining technique, ACM Int. Conf. Proceeding Ser., 2016, \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1145/2979779.2979866\u003c/span\u003e\u003cspan address=\"10.1145/2979779.2979866\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eud din NM, Dar RA, Rasool M, Assad A. Breast cancer detection using deep learning: Datasets, methods, and challenges ahead. Comput Biol Med. 2022;149:106073. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.compbiomed.2022.106073\u003c/span\u003e\u003cspan address=\"10.1016/j.compbiomed.2022.106073\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSalvi S, Kadam A. Breast Cancer Detection Using Deep learning and IoT Technologies. J Phys Conf Ser. 2021;1831(1). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1088/1742-6596/1831/1/012030\u003c/span\u003e\u003cspan address=\"10.1088/1742-6596/1831/1/012030\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhao P, Yoo I, Lancey R, Varghese E. Mobile applications for pain management: An app analysis for clinical usage. BMC Med Inf Decis Mak. 2019;19(1):1\u0026ndash;10. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1186/s12911-019-0827-7\u003c/span\u003e\u003cspan address=\"10.1186/s12911-019-0827-7\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAl Husaini MAS, Habaebi MH, Hameed SA, Islam MR, Gunawan TS. A Systematic Review of Breast Cancer Detection Using Thermography and Neural Networks. IEEE Access. 2020;8:208922\u0026ndash;37. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1109/ACCESS.2020.3038817\u003c/span\u003e\u003cspan address=\"10.1109/ACCESS.2020.3038817\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAl Husaini MAS, Habaebi MH, Islam MR, Gunawan TS. Self-detection of early breast cancer application with infrared camera and deep learning. Electron. 2021;10(20). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3390/electronics10202538\u003c/span\u003e\u003cspan address=\"10.3390/electronics10202538\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHiremath S, Karibasappa K. Neural Network Based Noise Identification in Digital Images. ACEEE Int J Netw Secur. 2011;02(03):3\u0026ndash;6.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSalami AM, Salih DM, Fadhil AF. Thermal Image Features and Noise Effects Analysis, in Proceedings of the 7th International Engineering Conference Research and Innovation Amid Global Pandemic, IEC 2021, Institute of Electrical and Electronics Engineers Inc., Feb. 2021, pp. 43\u0026ndash;47. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1109/IEC52205.2021.9476100\u003c/span\u003e\u003cspan address=\"10.1109/IEC52205.2021.9476100\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiu Q, Liu Z, Yong S, Jia K, Razmjooy N. Computer-aided breast cancer diagnosis based on image segmentation and interval analysis. Automatika. 2020;61(3):496\u0026ndash;506. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1080/00051144.2020.1785784\u003c/span\u003e\u003cspan address=\"10.1080/00051144.2020.1785784\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePriyadharsini MS, HIGH DENSITY NOISE FILTER METHOD FOR DENOISING MAMMOGRAM BREAST,. Data cquisition Process no November. 2023. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.5281/zenodo.776699\u003c/span\u003e\u003cspan address=\"10.5281/zenodo.776699\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMulaveesala R, Dua G. Non-invasive and non-ionizing depth resolved infra-red imaging for detection and evaluation of breast cancer: a \u0026hellip; Biomed. Phys Eng Express. 2018;2(5):1\u0026ndash;5. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1088/2057-1976/2/5/055004\u003c/span\u003e\u003cspan address=\"10.1088/2057-1976/2/5/055004\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYousefi B, SPAER, et al. Sparse deep convolutional autoencoder model to extract low dimensional imaging biomarkers for early detection of breast cancer using dynamic thermography. Appl Sci. 2021;11(7). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3390/app11073248\u003c/span\u003e\u003cspan address=\"10.3390/app11073248\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eEkici S, Jawzal H. Breast cancer diagnosis using thermography and convolutional neural networks, Med. Hypotheses, vol. 137, no. December 2019, p. 109542, 2020, \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.mehy.2019.109542\u003c/span\u003e\u003cspan address=\"10.1016/j.mehy.2019.109542\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKermani S, Samadzadehaghdam N, EtehadTavakol M, Optik, Stuttg.)., 126, 21, pp. 3288\u0026ndash;94, 2015, \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.ijleo.2015.08.007\u003c/span\u003e\u003cspan address=\"10.1016/j.ijleo.2015.08.007\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDalmia A, Kakileti ST, Manjunath G. Exploring Deep Learning Networks for Tumour Segmentation in Infrared Images 2, 14 th Quant. InfraRed Thermogr. Conf., pp. 1\u0026ndash;10, 2019, \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1080/17686733.2019.1619355\u003c/span\u003e\u003cspan address=\"10.1080/17686733.2019.1619355\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRoslidar R, et al. BreaCNet: A high-accuracy breast thermogram classifier based on mobile convolutional neural network. Math Biosci Eng. 2022;19(2):1304\u0026ndash;31. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3934/mbe.2022060\u003c/span\u003e\u003cspan address=\"10.3934/mbe.2022060\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGomathi P, Muniraj C, Periasamy PS. Digital infrared thermal imaging system based breast cancer diagnosis using 4D U-Net segmentation. Biomed Signal Process Control. 2023;85:104792. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.bspc.2023.104792\u003c/span\u003e\u003cspan address=\"10.1016/j.bspc.2023.104792\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAl Husaini MAS, Habaebi MH, Gunawan TS, Islam MR, Elsheikh EAA, Suliman FM. Thermal-based early breast cancer detection using inception V3, inception V4 and modified inception MV4. Neural Comput Appl. 2022;34(1):333\u0026ndash;48. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/s00521-021-06372-1\u003c/span\u003e\u003cspan address=\"10.1007/s00521-021-06372-1\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSzegedy C, Ioffe S, Vanhoucke V, Alemi A. Inception-v4, Inception-ResNet and the Impact of Residual Connections on Learning, Feb. 2016, [Online]. Available: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://arxiv.org/abs/1602.07261\u003c/span\u003e\u003cspan address=\"http://arxiv.org/abs/1602.07261\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSzegedy C, Ioffe S, Vanhoucke V, Alemi AA. Inception-v4, Inception-ResNet and the Impact of Residual Connections on Learning, in Proceedings of the Thirty-First AAAI Conference on Artificial Intelligence (AAAI-17) Inception-v4, 2017, pp. 4278\u0026ndash;4284. [Online]. Available: www.aaai.org.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGoodman JW. Some fundamental properties of speckle*. J Opt Soc Am. 1976;66(11):1145. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1364/josa.66.001145\u003c/span\u003e\u003cspan address=\"10.1364/josa.66.001145\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHiremath PS, P. T., and, Badiger S. Speckle Noise Reduction in Medical Ultrasound Images. Adv Break Ultrasound Imaging. 2013. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.5772/56519\u003c/span\u003e\u003cspan address=\"10.5772/56519\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSudha S, Suresh GR, Sukanesh R. Speckle Noise Reduction in Ultrasound Images by Wavelet Thresholding based on Weighted Variance. Int J Comput Theory Eng no April. 2009;7\u0026ndash;12. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.7763/ijcte.2009.v1.2\u003c/span\u003e\u003cspan address=\"10.7763/ijcte.2009.v1.2\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBuades A, Coll B, Morel JM. A non-local algorithm for image denoising, Proc. \u0026ndash;\u0026thinsp;2005 IEEE Comput. Soc. Conf. Comput. Vis. Pattern Recognition, CVPR 2005, vol. II, no. July, pp. 60\u0026ndash;65, 2005, \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1109/CVPR.2005.38\u003c/span\u003e\u003cspan address=\"10.1109/CVPR.2005.38\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRohit V, Ali J, A Comparative Study of Various Types of Image Noise and Efficient Noise Removal Techniques. Int J Adv Res Comput Sci Softw Eng. 2013;3(10):2277\u0026ndash;128.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSalmon J, Harmany Z, Deledalle CA, Willett R. Poisson noise reduction with non-local PCA. J Math Imaging Vis. 2014;48(2):279\u0026ndash;94. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/s10851-013-0435-6\u003c/span\u003e\u003cspan address=\"10.1007/s10851-013-0435-6\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"},{"header":"Tables","content":"\u003cp\u003eTable 1: Comparison between different Noise type and Deep learning models for early breast cancer detection.\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"657\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd width=\"15.068493150684931%\" valign=\"top\"\u003e\n \u003cp\u003eSource\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"28.767123287671232%\" valign=\"top\"\u003e\n \u003cp\u003eNoise type\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"32.42009132420091%\" valign=\"top\"\u003e\n \u003cp\u003eDeep learning model\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"23.744292237442924%\" valign=\"top\"\u003e\n \u003cp\u003eAccuracy\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"15.068493150684931%\" valign=\"top\"\u003e\n \u003cp\u003e[8]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"28.767123287671232%\" valign=\"top\"\u003e\n \u003cp\u003ewhite Gaussian noise\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"32.42009132420091%\" valign=\"top\"\u003e\n \u003cp\u003eNA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"23.744292237442924%\" valign=\"top\"\u003e\n \u003cp\u003eNA\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"15.068493150684931%\" valign=\"top\"\u003e\n \u003cp\u003e[9]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"28.767123287671232%\" valign=\"top\"\u003e\n \u003cp\u003eGaussian noise (3\u0026ndash;20% noise)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"32.42009132420091%\" valign=\"top\"\u003e\n \u003cp\u003esparse deep convolutional autoencoder (SPAER) model\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"23.744292237442924%\" valign=\"top\"\u003e\n \u003cp\u003e78.2%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"15.068493150684931%\" valign=\"top\"\u003e\n \u003cp\u003e[10]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"28.767123287671232%\" valign=\"top\"\u003e\n \u003cp\u003esalt and noise\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"32.42009132420091%\" valign=\"top\"\u003e\n \u003cp\u003eCNN\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"23.744292237442924%\" valign=\"top\"\u003e\n \u003cp\u003e98.95%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"15.068493150684931%\" valign=\"top\"\u003e\n \u003cp\u003e[11]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"28.767123287671232%\" valign=\"top\"\u003e\n \u003cp\u003ePoisson noise\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"32.42009132420091%\" valign=\"top\"\u003e\n \u003cp\u003eNA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"23.744292237442924%\" valign=\"top\"\u003e\n \u003cp\u003eNA\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"15.068493150684931%\" valign=\"top\"\u003e\n \u003cp\u003e[12]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"28.767123287671232%\" valign=\"top\"\u003e\n \u003cp\u003eGaussian noise\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"32.42009132420091%\" valign=\"top\"\u003e\n \u003cp\u003ecascaded CNN architecture\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"23.744292237442924%\" valign=\"top\"\u003e\n \u003cp\u003e92%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"15.068493150684931%\" valign=\"top\"\u003e\n \u003cp\u003e[13]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"28.767123287671232%\" valign=\"top\"\u003e\n \u003cp\u003eNormal noise\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"32.42009132420091%\" valign=\"top\"\u003e\n \u003cp\u003eBreaCNet\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"23.744292237442924%\" valign=\"top\"\u003e\n \u003cp\u003e100%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"15.068493150684931%\" valign=\"top\"\u003e\n \u003cp\u003e[14]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"28.767123287671232%\" valign=\"top\"\u003e\n \u003cp\u003especkle noise\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"32.42009132420091%\" valign=\"top\"\u003e\n \u003cp\u003eUnet\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"23.744292237442924%\" valign=\"top\"\u003e\n \u003cp\u003eimprovements with a 39.01%\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003eNA= Not Available\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eTable 2: Comparison of parameters in Figure 4 such as Mean Noise Value, Standard Deviation of Noise and Noise Value in dB.\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"575\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd width=\"16.202090592334496%\" valign=\"top\"\u003e\n \u003cp\u003eFigure 4 \u0026nbsp;\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"26.306620209059233%\" valign=\"top\"\u003e\n \u003cp\u003eMean Noise Value\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"33.44947735191638%\" valign=\"top\"\u003e\n \u003cp\u003eStandard Deviation of Noise\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"24.041811846689896%\" valign=\"top\"\u003e\n \u003cp\u003eNoise Value in dB\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"16.202090592334496%\" valign=\"top\"\u003e\n \u003cp\u003ea\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"26.306620209059233%\" valign=\"top\"\u003e\n \u003cp\u003e109.318\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"33.44947735191638%\" valign=\"top\"\u003e\n \u003cp\u003e112.828\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"24.041811846689896%\" valign=\"top\"\u003e\n \u003cp\u003e18.765\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"16.202090592334496%\" valign=\"top\"\u003e\n \u003cp\u003eb\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"26.306620209059233%\" valign=\"top\"\u003e\n \u003cp\u003e92.156\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"33.44947735191638%\" valign=\"top\"\u003e\n \u003cp\u003e119.703\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"24.041811846689896%\" valign=\"top\"\u003e\n \u003cp\u003e11.926\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"16.202090592334496%\" valign=\"top\"\u003e\n \u003cp\u003ec\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"26.306620209059233%\" valign=\"top\"\u003e\n \u003cp\u003e86.018\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"33.44947735191638%\" valign=\"top\"\u003e\n \u003cp\u003e117.159\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"24.041811846689896%\" valign=\"top\"\u003e\n \u003cp\u003e12.951\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cbr\u003e\u003c/p\u003e\n\u003cp\u003eTable 3: Average detection accuracy of different databases.\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"557\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd width=\"22.043010752688172%\"\u003e\n \u003cp\u003eEpoch\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"42.29390681003584%\"\u003e\n \u003cp\u003eGaussian Noise 0.05\u0026amp;0.02\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"13.620071684587813%\"\u003e\n \u003cp\u003eDMRA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"22.043010752688172%\"\u003e\n \u003cp\u003eDMRA+NOISE\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"22.043010752688172%\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"42.29390681003584%\"\u003e\n \u003cp\u003e96.06\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"13.620071684587813%\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"22.043010752688172%\"\u003e\n \u003cp\u003e99.91\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"22.043010752688172%\"\u003e\n \u003cp\u003e2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"42.29390681003584%\"\u003e\n \u003cp\u003e96.06\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"13.620071684587813%\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"22.043010752688172%\"\u003e\n \u003cp\u003e99.91\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"22.043010752688172%\"\u003e\n \u003cp\u003e3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"42.29390681003584%\"\u003e\n \u003cp\u003e96.96\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"13.620071684587813%\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"22.043010752688172%\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"22.043010752688172%\"\u003e\n \u003cp\u003e4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"42.29390681003584%\"\u003e\n \u003cp\u003e99.82\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"13.620071684587813%\"\u003e\n \u003cp\u003e99.82\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"22.043010752688172%\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"22.043010752688172%\"\u003e\n \u003cp\u003e5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"42.29390681003584%\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"13.620071684587813%\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"22.043010752688172%\"\u003e\n \u003cp\u003e99.82\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"22.043010752688172%\"\u003e\n \u003cp\u003e6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"42.29390681003584%\"\u003e\n \u003cp\u003e99.82\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"13.620071684587813%\"\u003e\n \u003cp\u003e99.92\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"22.043010752688172%\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"22.043010752688172%\"\u003e\n \u003cp\u003e7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"42.29390681003584%\"\u003e\n \u003cp\u003e99.64\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"13.620071684587813%\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"22.043010752688172%\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"22.043010752688172%\"\u003e\n \u003cp\u003e8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"42.29390681003584%\"\u003e\n \u003cp\u003e99.82\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"13.620071684587813%\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"22.043010752688172%\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"22.043010752688172%\"\u003e\n \u003cp\u003e9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"42.29390681003584%\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"13.620071684587813%\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"22.043010752688172%\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"22.043010752688172%\"\u003e\n \u003cp\u003e10\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"42.29390681003584%\"\u003e\n \u003cp\u003e99.82\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"13.620071684587813%\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"22.043010752688172%\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"22.043010752688172%\"\u003e\n \u003cp\u003eAverage\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"42.29390681003584%\"\u003e\n \u003cp\u003e98.8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"13.620071684587813%\"\u003e\n \u003cp\u003e99.974\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"22.043010752688172%\"\u003e\n \u003cp\u003e99.964\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cbr\u003e\u003c/p\u003e\n\u003cp\u003eTable 4: Average detection accuracy of different value in salt and pepper noise \u0026nbsp;\u0026nbsp;\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"474\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd width=\"30.78556263269639%\" valign=\"bottom\"\u003e\n \u003cp\u003eSetting\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.067940552016985%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.067940552016985%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"10.615711252653927%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.26\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.314225053078557%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.27\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"10.615711252653927%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.28\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.9171974522293%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.29\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"10.615711252653927%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.3\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"30.78556263269639%\" valign=\"bottom\"\u003e\n \u003cp\u003eaccuracy 1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.067940552016985%\" valign=\"bottom\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.067940552016985%\" valign=\"bottom\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"10.615711252653927%\" valign=\"bottom\"\u003e\n \u003cp\u003e99.96\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.314225053078557%\" valign=\"bottom\"\u003e\n \u003cp\u003e85\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"10.615711252653927%\" valign=\"bottom\"\u003e\n \u003cp\u003e62\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.9171974522293%\" valign=\"bottom\"\u003e\n \u003cp\u003e64\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"10.615711252653927%\" valign=\"bottom\"\u003e\n \u003cp\u003e18\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"30.78556263269639%\" valign=\"bottom\"\u003e\n \u003cp\u003eaccuracy 2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.067940552016985%\" valign=\"bottom\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.067940552016985%\" valign=\"bottom\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"10.615711252653927%\" valign=\"bottom\"\u003e\n \u003cp\u003e99.99\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.314225053078557%\" valign=\"bottom\"\u003e\n \u003cp\u003e99.9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"10.615711252653927%\" valign=\"bottom\"\u003e\n \u003cp\u003e74\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.9171974522293%\" valign=\"bottom\"\u003e\n \u003cp\u003e99\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"10.615711252653927%\" valign=\"bottom\"\u003e\n \u003cp\u003e99.9\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"30.78556263269639%\" valign=\"bottom\"\u003e\n \u003cp\u003eaccuracy 3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.067940552016985%\" valign=\"bottom\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.067940552016985%\" valign=\"bottom\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"10.615711252653927%\" valign=\"bottom\"\u003e\n \u003cp\u003e89.6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.314225053078557%\" valign=\"bottom\"\u003e\n \u003cp\u003e91.5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"10.615711252653927%\" valign=\"bottom\"\u003e\n \u003cp\u003e99.9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.9171974522293%\" valign=\"bottom\"\u003e\n \u003cp\u003e89\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"10.615711252653927%\" valign=\"bottom\"\u003e\n \u003cp\u003e28\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"30.78556263269639%\" valign=\"bottom\"\u003e\n \u003cp\u003eaccuracy 4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.067940552016985%\" valign=\"bottom\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.067940552016985%\" valign=\"bottom\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"10.615711252653927%\" valign=\"bottom\"\u003e\n \u003cp\u003e99.92\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.314225053078557%\" valign=\"bottom\"\u003e\n \u003cp\u003e98\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"10.615711252653927%\" valign=\"bottom\"\u003e\n \u003cp\u003e99.4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.9171974522293%\" valign=\"bottom\"\u003e\n \u003cp\u003e75\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"10.615711252653927%\" valign=\"bottom\"\u003e\n \u003cp\u003e52\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"30.78556263269639%\" valign=\"bottom\"\u003e\n \u003cp\u003eaccuracy 5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.067940552016985%\" valign=\"bottom\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.067940552016985%\" valign=\"bottom\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"10.615711252653927%\" valign=\"bottom\"\u003e\n \u003cp\u003e99.93\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.314225053078557%\" valign=\"bottom\"\u003e\n \u003cp\u003e99.98\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"10.615711252653927%\" valign=\"bottom\"\u003e\n \u003cp\u003e20\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.9171974522293%\" valign=\"bottom\"\u003e\n \u003cp\u003e81\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"10.615711252653927%\" valign=\"bottom\"\u003e\n \u003cp\u003e60\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"30.78556263269639%\" valign=\"bottom\"\u003e\n \u003cp\u003eaverage accuracy\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.067940552016985%\" valign=\"bottom\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.067940552016985%\" valign=\"bottom\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"10.615711252653927%\" valign=\"bottom\"\u003e\n \u003cp\u003e97.88\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.314225053078557%\" valign=\"bottom\"\u003e\n \u003cp\u003e94.876\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"10.615711252653927%\" valign=\"bottom\"\u003e\n \u003cp\u003e71.06\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.9171974522293%\" valign=\"bottom\"\u003e\n \u003cp\u003e81.6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"10.615711252653927%\" valign=\"bottom\"\u003e\n \u003cp\u003e51.58\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cbr\u003e\u003c/p\u003e\n\u003cp\u003eTable 5: Average detection accuracy of different value in Variance of multiplicative noise\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"514\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd width=\"25.78125%\" valign=\"bottom\"\u003e\n \u003cp\u003eSetting\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.1796875%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.1796875%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.421875%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.890625%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.890625%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.890625%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.765625%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.8\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"25.78125%\" valign=\"bottom\"\u003e\n \u003cp\u003eaccuracy 1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.1796875%\" valign=\"bottom\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.1796875%\" valign=\"bottom\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.421875%\" valign=\"bottom\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.890625%\" valign=\"bottom\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.890625%\" valign=\"bottom\"\u003e\n \u003cp\u003e99.996\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.890625%\" valign=\"bottom\"\u003e\n \u003cp\u003e99.8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.765625%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.6\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"25.78125%\" valign=\"bottom\"\u003e\n \u003cp\u003eaccuracy 2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.1796875%\" valign=\"bottom\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.1796875%\" valign=\"bottom\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.421875%\" valign=\"bottom\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.890625%\" valign=\"bottom\"\u003e\n \u003cp\u003e99.98\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.890625%\" valign=\"bottom\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.890625%\" valign=\"bottom\"\u003e\n \u003cp\u003e99.11\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.765625%\" valign=\"bottom\"\u003e\n \u003cp\u003e94\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"25.78125%\" valign=\"bottom\"\u003e\n \u003cp\u003eaccuracy 3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.1796875%\" valign=\"bottom\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.1796875%\" valign=\"bottom\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.421875%\" valign=\"bottom\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.890625%\" valign=\"bottom\"\u003e\n \u003cp\u003e99.999\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.890625%\" valign=\"bottom\"\u003e\n \u003cp\u003e99.999\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.890625%\" valign=\"bottom\"\u003e\n \u003cp\u003e99.999\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.765625%\" valign=\"bottom\"\u003e\n \u003cp\u003e30\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"25.78125%\" valign=\"bottom\"\u003e\n \u003cp\u003eaccuracy 4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.1796875%\" valign=\"bottom\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.1796875%\" valign=\"bottom\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.421875%\" valign=\"bottom\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.890625%\" valign=\"bottom\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.890625%\" valign=\"bottom\"\u003e\n \u003cp\u003e99.998\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.890625%\" valign=\"bottom\"\u003e\n \u003cp\u003e49.6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.765625%\" valign=\"bottom\"\u003e\n \u003cp\u003e8\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"25.78125%\" valign=\"bottom\"\u003e\n \u003cp\u003eaccuracy 5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.1796875%\" valign=\"bottom\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.1796875%\" valign=\"bottom\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.421875%\" valign=\"bottom\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.890625%\" valign=\"bottom\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.890625%\" valign=\"bottom\"\u003e\n \u003cp\u003e99.52\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.890625%\" valign=\"bottom\"\u003e\n \u003cp\u003e99.77\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.765625%\" valign=\"bottom\"\u003e\n \u003cp\u003e86.7\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"25.78125%\" valign=\"bottom\"\u003e\n \u003cp\u003eaverage accuracy\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.1796875%\" valign=\"bottom\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.1796875%\" valign=\"bottom\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.421875%\" valign=\"bottom\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.890625%\" valign=\"bottom\"\u003e\n \u003cp\u003e99.9958\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.890625%\" valign=\"bottom\"\u003e\n \u003cp\u003e99.9026\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.890625%\" valign=\"bottom\"\u003e\n \u003cp\u003e89.6558\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.765625%\" valign=\"bottom\"\u003e\n \u003cp\u003e43.86\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cbr\u003e\u003c/p\u003e\n\u003cp\u003eTable 6: Average detection accuracy of different value in speckle noise\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"522\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd width=\"25.384615384615383%\" valign=\"bottom\"\u003e\n \u003cp\u003eSetting\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.038461538461538%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.038461538461538%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.3076923076923075%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.692307692307692%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.692307692307692%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.692307692307692%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.153846153846153%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.8\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"25.384615384615383%\" valign=\"bottom\"\u003e\n \u003cp\u003eaccuracy 1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.038461538461538%\" valign=\"bottom\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.038461538461538%\" valign=\"bottom\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.3076923076923075%\" valign=\"bottom\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.692307692307692%\" valign=\"bottom\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.692307692307692%\" valign=\"bottom\"\u003e\n \u003cp\u003e99.998\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.692307692307692%\" valign=\"bottom\"\u003e\n \u003cp\u003e95.4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.153846153846153%\" valign=\"bottom\"\u003e\n \u003cp\u003e97.77\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"25.384615384615383%\" valign=\"bottom\"\u003e\n \u003cp\u003eaccuracy 2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.038461538461538%\" valign=\"bottom\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.038461538461538%\" valign=\"bottom\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.3076923076923075%\" valign=\"bottom\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.692307692307692%\" valign=\"bottom\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.692307692307692%\" valign=\"bottom\"\u003e\n \u003cp\u003e99.994\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.692307692307692%\" valign=\"bottom\"\u003e\n \u003cp\u003e99.55\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.153846153846153%\" valign=\"bottom\"\u003e\n \u003cp\u003e44.5\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"25.384615384615383%\" valign=\"bottom\"\u003e\n \u003cp\u003eaccuracy 3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.038461538461538%\" valign=\"bottom\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.038461538461538%\" valign=\"bottom\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.3076923076923075%\" valign=\"bottom\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.692307692307692%\" valign=\"bottom\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.692307692307692%\" valign=\"bottom\"\u003e\n \u003cp\u003e99.999\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.692307692307692%\" valign=\"bottom\"\u003e\n \u003cp\u003e99.84\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.153846153846153%\" valign=\"bottom\"\u003e\n \u003cp\u003e3.3\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"25.384615384615383%\" valign=\"bottom\"\u003e\n \u003cp\u003eaccuracy 4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.038461538461538%\" valign=\"bottom\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.038461538461538%\" valign=\"bottom\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.3076923076923075%\" valign=\"bottom\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.692307692307692%\" valign=\"bottom\"\u003e\n \u003cp\u003e99.99\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.692307692307692%\" valign=\"bottom\"\u003e\n \u003cp\u003e98.8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.692307692307692%\" valign=\"bottom\"\u003e\n \u003cp\u003e99.99\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.153846153846153%\" valign=\"bottom\"\u003e\n \u003cp\u003e69\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"25.384615384615383%\" valign=\"bottom\"\u003e\n \u003cp\u003eaccuracy 5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.038461538461538%\" valign=\"bottom\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.038461538461538%\" valign=\"bottom\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.3076923076923075%\" valign=\"bottom\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.692307692307692%\" valign=\"bottom\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.692307692307692%\" valign=\"bottom\"\u003e\n \u003cp\u003e99.968\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.692307692307692%\" valign=\"bottom\"\u003e\n \u003cp\u003e99.92\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.153846153846153%\" valign=\"bottom\"\u003e\n \u003cp\u003e96\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"25.384615384615383%\" valign=\"bottom\"\u003e\n \u003cp\u003eaverage accuracy\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.038461538461538%\" valign=\"bottom\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.038461538461538%\" valign=\"bottom\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.3076923076923075%\" valign=\"bottom\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.692307692307692%\" valign=\"bottom\"\u003e\n \u003cp\u003e99.998\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.692307692307692%\" valign=\"bottom\"\u003e\n \u003cp\u003e99.7518\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.692307692307692%\" valign=\"bottom\"\u003e\n \u003cp\u003e98.94\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.153846153846153%\" valign=\"bottom\"\u003e\n \u003cp\u003e62.114\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cbr\u003e\u003c/p\u003e\n\u003cp\u003eTable 7: Evaluation parameters for different type of database\u0026nbsp;\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"652\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd width=\"18.364197530864196%\" valign=\"bottom\"\u003e\n \u003cp\u003e\u0026nbsp;Type of database\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.716049382716049%\" valign=\"bottom\"\u003e\n \u003cp\u003eAccu\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.716049382716049%\" valign=\"bottom\"\u003e\n \u003cp\u003eSens\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.716049382716049%\" valign=\"bottom\"\u003e\n \u003cp\u003eSpec\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.716049382716049%\" valign=\"bottom\"\u003e\n \u003cp\u003eP\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.716049382716049%\" valign=\"bottom\"\u003e\n \u003cp\u003eNPV\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.716049382716049%\" valign=\"bottom\"\u003e\n \u003cp\u003eFPR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.716049382716049%\" valign=\"bottom\"\u003e\n \u003cp\u003eFNR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.716049382716049%\" valign=\"bottom\"\u003e\n \u003cp\u003eLRN\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.716049382716049%\" valign=\"bottom\"\u003e\n \u003cp\u003eAUC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.716049382716049%\" valign=\"bottom\"\u003e\n \u003cp\u003eEER\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"4.4753086419753085%\" valign=\"bottom\"\u003e\n \u003cp\u003eF1\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"18.364197530864196%\" valign=\"bottom\"\u003e\n \u003cp\u003eGaussian Noise 0.05\u0026amp;0.02\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.716049382716049%\" valign=\"bottom\"\u003e\n \u003cp\u003e96.06\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.716049382716049%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.970\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.716049382716049%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.952\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.716049382716049%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.946\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.716049382716049%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.974\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.716049382716049%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.048\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.716049382716049%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.030\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.716049382716049%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.031\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.716049382716049%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.961\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.716049382716049%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.039\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"4.4753086419753085%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.958\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"18.364197530864196%\" valign=\"bottom\"\u003e\n \u003cp\u003eDMRA+NOISE\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.716049382716049%\" valign=\"bottom\"\u003e\n \u003cp\u003e99.82\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.716049382716049%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.996\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.716049382716049%\" valign=\"bottom\"\u003e\n \u003cp\u003e1.000\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.716049382716049%\" valign=\"bottom\"\u003e\n \u003cp\u003e1.000\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.716049382716049%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.997\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.716049382716049%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.000\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.716049382716049%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.004\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.716049382716049%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.004\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.716049382716049%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.998\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.716049382716049%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.002\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"4.4753086419753085%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.998\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"18.364197530864196%\" valign=\"bottom\"\u003e\n \u003cp\u003eDRMA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.716049382716049%\" valign=\"bottom\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.716049382716049%\" valign=\"bottom\"\u003e\n \u003cp\u003e1.000\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.716049382716049%\" valign=\"bottom\"\u003e\n \u003cp\u003e1.000\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.716049382716049%\" valign=\"bottom\"\u003e\n \u003cp\u003e1.000\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.716049382716049%\" valign=\"bottom\"\u003e\n \u003cp\u003e1.000\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.716049382716049%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.000\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.716049382716049%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.000\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.716049382716049%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.000\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.716049382716049%\" valign=\"bottom\"\u003e\n \u003cp\u003e1.000\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.716049382716049%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.000\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"4.4753086419753085%\" valign=\"bottom\"\u003e\n \u003cp\u003e1.000\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eTable 8: Average detection accuracy of different value Poisson noise\u0026nbsp;\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"675\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd width=\"16.59259259259259%\"\u003e\n \u003cp\u003ePoisson noise Setting\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"5.62962962962963%\"\u003e\n \u003cp\u003e0.2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"6.518518518518518%\"\u003e\n \u003cp\u003eSRN\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"5.62962962962963%\"\u003e\n \u003cp\u003e0.3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"6.518518518518518%\"\u003e\n \u003cp\u003eSRN\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"5.037037037037037%\"\u003e\n \u003cp\u003e0.4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"6.37037037037037%\"\u003e\n \u003cp\u003eSRN\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"5.62962962962963%\"\u003e\n \u003cp\u003e0.5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"6.37037037037037%\"\u003e\n \u003cp\u003eSRN\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"5.333333333333333%\"\u003e\n \u003cp\u003e0.6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"6.37037037037037%\"\u003e\n \u003cp\u003eSRN\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"5.62962962962963%\"\u003e\n \u003cp\u003e0.7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"6.37037037037037%\"\u003e\n \u003cp\u003eSRN\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"5.62962962962963%\"\u003e\n \u003cp\u003e0.8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"6.37037037037037%\"\u003e\n \u003cp\u003eSRN\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"16.59259259259259%\" valign=\"bottom\"\u003e\n \u003cp\u003eaccuracy 1 \u0026nbsp; %\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"5.62962962962963%\" valign=\"bottom\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"6.518518518518518%\" rowspan=\"5\"\u003e\n \u003cp\u003e13.98 dB\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"5.62962962962963%\" valign=\"bottom\"\u003e\n \u003cp\u003e99.9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"6.518518518518518%\" rowspan=\"5\"\u003e\n \u003cp\u003e10.46 dB\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"5.037037037037037%\" valign=\"bottom\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"6.37037037037037%\" rowspan=\"5\"\u003e\n \u003cp\u003e7.96 dB\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"5.62962962962963%\" valign=\"bottom\"\u003e\n \u003cp\u003e99.9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"6.37037037037037%\" rowspan=\"5\"\u003e\n \u003cp\u003e6.02 dB\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"5.333333333333333%\" valign=\"bottom\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"6.37037037037037%\" rowspan=\"5\"\u003e\n \u003cp\u003e4.44 dB\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"5.62962962962963%\" valign=\"bottom\"\u003e\n \u003cp\u003e99.9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"6.37037037037037%\" rowspan=\"5\"\u003e\n \u003cp\u003e3.10 dB\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"5.62962962962963%\" valign=\"bottom\"\u003e\n \u003cp\u003e99.9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"6.37037037037037%\" rowspan=\"5\"\u003e\n \u003cp\u003e1.94 dB\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"30.107526881720432%\" valign=\"bottom\"\u003e\n \u003cp\u003eaccuracy 2 \u0026nbsp; %\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"10.21505376344086%\" valign=\"bottom\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"10.21505376344086%\" valign=\"bottom\"\u003e\n \u003cp\u003e99.9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.13978494623656%\" valign=\"bottom\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"10.21505376344086%\" valign=\"bottom\"\u003e\n \u003cp\u003e99.9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.67741935483871%\" valign=\"bottom\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"10.21505376344086%\" valign=\"bottom\"\u003e\n \u003cp\u003e99.9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"10.21505376344086%\" valign=\"bottom\"\u003e\n \u003cp\u003e99.9\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"30.107526881720432%\" valign=\"bottom\"\u003e\n \u003cp\u003eaccuracy 3 \u0026nbsp; %\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"10.21505376344086%\" valign=\"bottom\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"10.21505376344086%\" valign=\"bottom\"\u003e\n \u003cp\u003e99.9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.13978494623656%\" valign=\"bottom\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"10.21505376344086%\" valign=\"bottom\"\u003e\n \u003cp\u003e99.9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.67741935483871%\" valign=\"bottom\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"10.21505376344086%\" valign=\"bottom\"\u003e\n \u003cp\u003e99.9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"10.21505376344086%\" valign=\"bottom\"\u003e\n \u003cp\u003e99.9\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"30.107526881720432%\" valign=\"bottom\"\u003e\n \u003cp\u003eaccuracy 4 \u0026nbsp; %\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"10.21505376344086%\" valign=\"bottom\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"10.21505376344086%\" valign=\"bottom\"\u003e\n \u003cp\u003e99.9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.13978494623656%\" valign=\"bottom\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"10.21505376344086%\" valign=\"bottom\"\u003e\n \u003cp\u003e99.9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.67741935483871%\" valign=\"bottom\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"10.21505376344086%\" valign=\"bottom\"\u003e\n \u003cp\u003e99.9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"10.21505376344086%\" valign=\"bottom\"\u003e\n \u003cp\u003e99.9\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"30.107526881720432%\" valign=\"bottom\"\u003e\n \u003cp\u003eaccuracy 5 \u0026nbsp; %\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"10.21505376344086%\" valign=\"bottom\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"10.21505376344086%\" valign=\"bottom\"\u003e\n \u003cp\u003e99.9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.13978494623656%\" valign=\"bottom\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"10.21505376344086%\" valign=\"bottom\"\u003e\n \u003cp\u003e99.9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.67741935483871%\" valign=\"bottom\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"10.21505376344086%\" valign=\"bottom\"\u003e\n \u003cp\u003e99.9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"10.21505376344086%\" valign=\"bottom\"\u003e\n \u003cp\u003e99.9\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"16.59259259259259%\"\u003e\n \u003cp\u003eaverage accuracy \u0026nbsp; \u0026nbsp;%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"5.62962962962963%\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"6.518518518518518%\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"5.62962962962963%\"\u003e\n \u003cp\u003e99.9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"6.518518518518518%\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"5.037037037037037%\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"6.37037037037037%\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"5.62962962962963%\"\u003e\n \u003cp\u003e99.9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"6.37037037037037%\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"5.333333333333333%\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"6.37037037037037%\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"5.62962962962963%\"\u003e\n \u003cp\u003e99.9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"6.37037037037037%\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"5.62962962962963%\"\u003e\n \u003cp\u003e99.9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"6.37037037037037%\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cbr\u003e\u003c/p\u003e\n\u003cp\u003eTable 9: Confusion matrix for Inception v3, Inception v4 and Inception mv4\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"539\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd width=\"29.87012987012987%\" valign=\"bottom\"\u003e\n \u003cp\u003eDeep learning Models\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"20.964749536178108%\" valign=\"bottom\"\u003e\n \u003cp\u003eInception v3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"19.29499072356215%\" valign=\"bottom\"\u003e\n \u003cp\u003eInception v4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"29.87012987012987%\" valign=\"bottom\"\u003e\n \u003cp\u003eInception mv4\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"29.87012987012987%\" valign=\"bottom\"\u003e\n \u003cp\u003eTrue Positive (TP)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"20.964749536178108%\" valign=\"bottom\"\u003e\n \u003cp\u003e255.000\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"19.29499072356215%\" valign=\"bottom\"\u003e\n \u003cp\u003e171.000\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"29.87012987012987%\" valign=\"bottom\"\u003e\n \u003cp\u003e171.000\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"29.87012987012987%\" valign=\"bottom\"\u003e\n \u003cp\u003e\u0026nbsp;True Negative (TN)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"20.964749536178108%\" valign=\"bottom\"\u003e\n \u003cp\u003e289.000\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"19.29499072356215%\" valign=\"bottom\"\u003e\n \u003cp\u003e203.000\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"29.87012987012987%\" valign=\"bottom\"\u003e\n \u003cp\u003e203.000\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"29.87012987012987%\" valign=\"bottom\"\u003e\n \u003cp\u003eFalse Positive (FP)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"20.964749536178108%\" valign=\"bottom\"\u003e\n \u003cp\u003e3.000\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"19.29499072356215%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.000\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"29.87012987012987%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.000\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"29.87012987012987%\" valign=\"bottom\"\u003e\n \u003cp\u003eFalse Negative (FN)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"20.964749536178108%\" valign=\"bottom\"\u003e\n \u003cp\u003e15.000\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"19.29499072356215%\" valign=\"bottom\"\u003e\n \u003cp\u003e1.000\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"29.87012987012987%\" valign=\"bottom\"\u003e\n \u003cp\u003e1.000\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Breast Cancer, Thermal Image, Gaussian Noise, Deep Learning","lastPublishedDoi":"10.21203/rs.3.rs-3978583/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-3978583/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eBreast cancer remains a leading cause of mortality among women globally. There were techniques that have been developed to enhance early detection, among which thermal imaging has emerged as a promising modality capable of identifying potential signs of breast cancer in its early stages. In addition, Thermal images provide valuable pixel-level information by capturing temperature variations between healthy and cancerous tissues. However, the susceptibility of these thermal images to noise poses a challenge to the diagnostic accuracy in early stages. This research aims to assess the influence of various types of noise on performance of recently developed different deep learning models designed for early breast cancer detection. In addition, a comprehensive analysis was conducted using a substantial database to assess the impact of noise on the models' efficacy. Also, encompasses different categories of noise characterized by distinct mean and variance values ranging from 0.01 to 0.09. The findings reveal that the introduction of different types of noise, albeit within a small range of mean and variance values, adversely affects the performance of deep learning models. It shows that these filters play a pivotal role in enhancing the accuracy of classification. Moreover, the results show that salt and pepper noise, varied between 0.1 and 0.3, significantly impacted the accuracy of inception MV4, reducing it from 100\u0026ndash;51.58%, without adding filters in pre-processing. Additionally, the introduction of variance in multiplicative noise from 0.2 to 0.8, demonstrated an effect on classification accuracy only at noise levels of 0.7 (89%) and 0.8 (43%). Moreover, the results show that performance metrics for proposed method were accuracy of 99.82%, sensitivity of 0.996, specificity of 1, precision of 1, NPV of 0.997, FNR of 0.004, LRN of 0.004, AUC of 0.998, EER of 0.002, and F1 score of 0.998, but FPR of 0. In conclusion, findings underscore the significance of refining both noise mitigation strategies and preprocessing techniques to advance reliability and accuracy of thermal imaging as a diagnostic tool in breast cancer detection in early stages.\u003c/p\u003e","manuscriptTitle":"Evaluating the Effect of Noisy Thermal Images On the Detection of Early Breast Cancer Using Deep Learning","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-03-11 19:10:10","doi":"10.21203/rs.3.rs-3978583/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"867f2200-ee88-452c-b8fe-d557bda88e39","owner":[],"postedDate":"March 11th, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2024-08-29T05:31:54+00:00","versionOfRecord":[],"versionCreatedAt":"2024-03-11 19:10:10","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-3978583","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-3978583","identity":"rs-3978583","version":["v1"]},"buildId":"7rjqhiLT3MXkJMwkYKINL","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}