A Data-Driven Approach to Detecting Lung Cancer with Smart VGG19 Algorithm

A Data-Driven Approach to Detecting Lung Cancer with Smart VGG19 Algorithm | 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 Article A Data-Driven Approach to Detecting Lung Cancer with Smart VGG19 Algorithm Vinayak Vinayak This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7488263/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 One of the main causes of cancer-related deaths, lung cancer, requires early detection in order to be effectively treated. Our proposal is an automated method that evaluates CT data using machine learning to assist radiologists in accurately identifying troublesome lung nodules. The method focuses on nodule features via pre-processing CT images. Annotated CT scans are mostly used to train machine learning models, like CNNs, to recognize patterns of benign and cancerous nodules. To find the best effective model for detection, several CNN architectures are tested, such as ResNet, DarkNet, and EfficientNet. We present the MMT-VIN Basic model, which is based on the VGG-19 methodology. With an accuracy of almost 97.58%, the suggested lung cancer detection method outperforms the techniques it was compared to. Biological sciences/Cancer Biological sciences/Computational biology and bioinformatics Physical sciences/Engineering Physical sciences/Mathematics and computing Health sciences/Medical research CNN architectures CT images Lung cancer detection Machine learning MMT-VIN basic model Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 I. INTRODUCTION The primary cause of cancer-related death worldwide is still lung cancer. Risk factors for lung cancer include radiation therapy, alcohol, radon gas, asbestos, secondhand smoke, smoking, radiation therapy, and a family history of the disease [ 1 ]. Survival chances for lung cancer can be considerably increased by early detection. Image processing, machine learning, or hybrid approaches from the literature are used by researchers to detect lung cancer by Naive Bayes (NB), Convolutional Neural Networks (CNN), Support Vector Machines (SVM), Decision Trees (DT), Back Propagation Networks (BPN), Logistic Regression (LR), and K-Nearest Neighbors (KNN) [ 2 ]. Compared to other neural networks, CNNs have fewer parameters thanks to their pooling layers and local receptive fields. They decrease false positives and do away with manual nodule signifiers [ 3 ]. Chest radiography (CXR), the most widely used imaging diagnostic method, can be used to diagnose disorders of the chest, including emphysema, pneumonia, pneumoconiosis, lung cancer, and tuberculosis. It is the most widely used, least expensive, and least radioactive diagnostic method that can identify undiscovered disease alterations [ 4 ].Computed tomography (CT) has a high sensitivity when compared to computed radiography (CR), and early CT screening has a remarkable influence on the diagnosis of lung cancer [ 5 ]. In [ 6 ], a deep learning method for predicting the kind of lung cancer from CT scans was presented. This model classifies malignant tumors using densely linked convolutional networks (DenseNet), and it uses adaptive boosting (adaboost) to aggregate classification results for better performance. With 89.95% accuracy, the DenseNet model beat the VGG16, AlexNet, ResNet, and DenseNet without a boost. In another study, the authors of [ 7 ] presented a model with a denoising and detection component termed DFD-Net ("denoising first" two-path convolutional neural network). The DR-Net model, a residual learning denoising model, is used to initially remove the noise during the pre-processing phase. To detect lung cancer, the denoised images from the DR-Net model are subsequently fed into a two-path convolutional neural network. Discriminant correlation analysis enhances this model's performance by include more representative features, as opposed to the conventional feature concatenation method, which concatenates two sets of data from different CNN layers. Lastly, a retraining approach is suggested to get rid of problems related to the picture label imbalance. An improved CT scan method for lung nodule detection of CAD was presented in another study [ 8 ]. This method segments lung nodules from CT scans using a fuzzy-based algorithm. Fuzzy soft membership functions are used to automate segmentation. Lung nodules are reliably classified as benign or malignant by fuzzy neural classifiers. [ 9 ] Presented a poor classifier, and multiple VGG16-T neural networks are trained using the boosting technique. Accuracy is increased by combining weak classifier VGG16-T networks. II. Literature Review Because machine learning algorithms make it possible to analyze medical data more quickly and accurately, they are revolutionizing healthcare diagnostics. These algorithms can help clinicians make well-informed judgments by spotting trends and anomalies in big datasets [ 10 ]. Predictive analytics for patient outcomes, individualized treatment strategies, and early disease identification are some examples of applications. Healthcare systems can increase the efficiency of patient care and improve diagnostic accuracy by utilizing machine learning [ 11 ]. Researchers have used machine learning and deep learning for categorisation and prediction across imaging and signal modalities. Machine learning algorithms are being used more and more in medical diagnostic systems to forecast lung conditions [ 12 ]. DarkNet-19 is a deep learning network that uses YOLO object detection [ 13 ]. There are five pooling layers and 19 convolutional layers in DarkNet-19. Similar to VGG models, the DarkNet-19 model has twice as many channels but employs 3x3 filters following convolutional layers. Batch normalization is another technique used by the DarkNet-19 model to stabilize training, accelerate convergence, and simplify model operation. In another work, the DarkNet-19 model, a deep learning model, was used to train the image classes from scratch [ 14 ]. Using the feature set obtained from the DarkNet-19 model, the unsuccessful features were selected using the optimisation techniques of Manta Ray Foraging and Equilibrium. Complementary rule insets are feature sets that remain efficient after the inefficient features have been isolated from the rest of the set.The two employed the following successful characteristics: integrating optimization methods with the SVM classification approach. The EfficientNet model, which uses a technique known as compound scaling to modify all dimensions, is examined by the authors of [ 17 ]. In this strategy, FLOPs and accuracy are optimized by a multi-objective architectural search. EfficientNet employs a search space and the ACC (m) multiplyby [FLOPS (m)/T]w metric for optimization. FLOPS and ACC and symbolize the accuracy and FLOPs of model "m," while T and wstand for the preferred hyper-parameters and FLOPs. These elements are essential for striking a balance between computing efficiency and accuracy. This network consists of several convolutional layers with varying kernel sizes. The first input frame is 224 × 224 × 3 and has three color channels (red, green, and blue). As layers develop, width increases to boost accuracy and resolution falls to condense feature maps. This method gathers important input frame characteristics. Kernels are 112 pixels wide in 2nd layer and 64 pixels wide in the subsequent convolutional layer. To capture the most noticeable features, the final layer employs up to 2,560 kernels at a decreased resolution of 7 × 7. For classification, SoftMax layers, encoding, and max pooling are added. Google unveiled the EfficientNet [ 18 ], a collection of convolutional neural networks (CNNs) developed using an automated neural network architecture search. CNNs are designed more efficiently because to this automated method, which produces models with improved performance and lower processing costs. All of the models in the EfficientNet family (EfficientNet b0 through b7) are designed to operate best with particular image dimensions in order to balance efficiency and performance. For this study, EfficientNet b3 was chosen since its input size is similar to that of the InceptionResNetV2 model. According to [ 20 ], lung nodules in low-dose computed tomography (LDCT) images can be found early by employing computer-aided detection (CADe). To improve the contrast of low-dose images, our proposed method starts with data processing. Analysing deep learning architectures such as VGG16, VGG19, and Alex produces succinct features. On the dataset, all transfer learning techniques—including VGG 16, VGG 19, and Xception with a 20-epoch structure—perform admirably. Preprocessing facilitates the creation of a reliable model and speeds up model integration, allowing for the prediction of future events. The accuracy of VGG 16 is 98.83 percent by the twentieth epoch, 98.05 percent by VGG 19, and 97.4 percent by Xception. GoramMufarah M. Alshmrani et al. [ 22 ] employed a fully connected network for classification, three CNN blocks for feature extraction, and a pre-trained VGG19 model. The use of deep learning (DL) algorithms to analyse CT scan data has lately proved essential in enhancing lung cancer detection, especially in the categorisation of malignancy. Deep learning methods in medical imaging fall into two main categories: 2D and 3D models. The channel is the colour channel number, which is typically one for greyscale radiological images and three for natural photos in RGB scale. Two-dimensional deep learning models use two-dimensional images in a (channel, width, height) format [ 23 ]. In terms of lung cancer detection, machine learning holds enormous potential for enabling early identification and potentially improving patient outcomes. In this area, feature extraction remains a major challenge. Combining the most relevant features can increase detection accuracy even more. This work employs a hybrid feature extraction method that combines Gray-level co-occurrence matrix (GLCM) with Haralick and autoencoder features. Afterwards, supervised machine learning methods were trained using these attributes as input. GLCM with autoencoder, Haralick, and autoencoder features yielded an accuracy of 99.89% for SVM polynomial, but flawless performance metrics were achieved by SVM Gaussian and SVM Radial Base Function (RBF) [ 24 ]. A label-free detection approach for noninvasive biofluids enables early-stage cancer diagnosis and rapid on-site disease screening by looking at metabolic alterations. Here, we develop three-dimensional plasmonic hexaplex nanostructures that are deposited onto a 3D-PHP paper substrate. This adaptable and extremely absorbent 3D-PHP sensor is used in conjunction with a commercial saliva collection tube to create an efficient on-site sensing platform for lung cancer screening by detecting human saliva using surface-enhanced Raman scattering (SERS). The enhanced interface between the multispikehexaplex-shaped gold nanostructure and saliva viscosity enables efficient sampling and SERS enhancement. Through the analysis of patient salivary samples, the 3D-PHP sensor effectively identifies and diagnoses lung cancer. A logistic regression-based machine learning algorithm successfully differentiates between benign and malignant patients, demonstrating high clinical sensitivity and specificity [ 25 ]. Deep learning has emerged as a powerful tool for medical image analysis and diagnosis, according to the author of [ 26 ], who also noted that deep learning performs well on tasks. This literature review summarises recent studies on deep learning techniques for lung cancer screening and diagnosis. Significant advancements, limitations, and potential future directions in the state-of-the-art in deep learning for lung cancer diagnosis are highlighted in this work. We prioritised studies that provided a comprehensive view of the area by utilising significant public datasets such as LUNA16, JSRT, and LIDC. We focus on deep learning architectures which include 2D and 3D convolutional neural networks (CNNs), vision transformers (ViT), dual-path networks, and natural language processing (NLP). A comprehensive Systematic Literature Review (SLR) for lung cancer research is conducted by this study [ 27 ] using deep learning techniques. It provides a detailed analysis of the methodology, cutting-edge developments, quality assessments, and customised deep learning approaches. While there are many deep learning methods for classifying lung cancer, this study focusses mostly on the most popular method, the Convolutional Neural Network (CNN). The highest accuracy is achieved by CNN due to its multi-layer architecture, automated weight learning, and capacity to transfer local weights. CNN consistently shows the highest accuracy when performance parameters like precision, accuracy, specificity, sensitivity, and AUC are shown for multiple methods. The findings show that DCNN boosts lung cancer detection and classification greatly. The models that classified nodules into six groups using convolutional neural networks and support vector machines had an area under the curve of 0.89/0.87 between atypical adenomatous hyperplasia + adenocarcinoma in situ and minimally invasive adenocarcinoma and invasive adenocarcinoma, 0.87/0.86 between minimally invasive adenocarcinoma and invasive adenocarcinoma, 0.76/0.72 between atypical adenomatous hyperplasia + adenocarcinoma in situ and minimally invasive adenocarcinoma, and 0.93/0.92 between atypical adenomatous hyperplasia + adenocarcinoma in situ and minimally invasive adenocarcinoma. For the categorisation of adenocarcinoma in situ, minimally invasive versus invasive cancer, and benign versus atypical adenomatous hyperplasia, the support vector machine/convolutional neural network models yielded a micro-average area under the curve of 0.93/0.94 [ 28 ]. Additional useful information about a lung cancer diagnosis can be obtained from images from computer tomography (CT) scans. Using CT scan input images, several Machine Learning (ML) and Deep Learning (DL) algorithms are created to enhance diagnosis and treatment procedures. But the hardest aspect of research is still creating a sophisticated and accurate system. This paper proposes a novel classification technique called Deep Fused Features-Based Cat-Optimized Networks (DFF-CON), which is based on the concepts of fused features and optimised networks. In the proposed structure, Deep Convolutional Neural Networks (DCNN) are employed to enhance the classification maps and ultimately decrease the probability of an overfitting issue [ 29 ]. Table 1 Analysis and Comparison of Different CNN Models Author Name Approach Advantages Limitations J. Redmonet. al. [ 13 ], M. Toğaçar[ 14 ] Deep learning using DarkNet-19 with YOLO object detection and feature optimization (MRFO + Equilibrium) Efficient feature extraction, faster convergence, improved training stability Computational complexity; requires deep learning expertise K. Muhammadet. al.[ 17 ], A. Langet. al.[ 18 ] Compound scaling in CNNs; multi-objective architecture search for model optimization High accuracy with low computational cost; scalable across devices Architecture complexity; requires careful tuning A. Elnakibet. al.[ 20 ] CADe system using VGG16, VGG19, and AlexNet on LDCT scans Early-stage detection, improved image contrast through preprocessing Relatively shallow networks compared to newer models M. Humayunet. al.[ 21 ] Non-invasive diagnostic strategy with transfer learning (VGG16, VGG19, Xception) Fewer parameters, robust on varying dataset sizes, suitable for clinical evaluation May still require high-end computation for training G. M. M. Alshmraniet. al. [ 22 ] Three CNN blocks for feature extraction, a fully linked network for classification, and a pre-trained VGG19 Effective feature learning, structured model pipeline Requires high-quality pre-training and labeled datasets This work aims to create a trustworthy algorithm for the diagnosis of lung cancer that provides improved sensitivity and specificity while retaining low time complexity, motivated by current research initiatives. The work's specifics and the suggested methodology are described in the following sections. III. PROPOSED METHODOLOGY The suggested method for detecting lung cancer is divided into several stages, each of which is described in turn. An overview of the process workflow is introduced initially.After that data set collection and data preprocessing is discussed. Model architecture selection and model training explained later. Model evaluation, deployment discussed next. At the end results and discussion and conclusion given in this paper. A. Overview of the Process Workflow Classifying lung cancer using a CNN based on the MMT-Vin model with VGG-19 architecture involves several stages, including dataset collection, image preprocessing, model training, and others. Figures 1 and 2 provide a concise overview of the proposed method. B. Dataset Collection A large collection of lung cancer photos, including both benign and malignant examples, is shown in this section. Accurate labeling and diversity in the dataset are critical for the CNN model to learn efficiently. During the data collection stage of this study, the lung cancer histology photos were taken from Kaggle. Upon unzipping, we discovered that there are 515 pictures of lung cancer in the collection. After augmentation, we were able to locate 4,250 images. C. Data Pre-processing First, we convert BGR (blue, green, and red) photos to grayscale. To generate a single intensity value for every pixel, the conversion employs distinct weights for every channel. The often used formula is gray = 0.299×R + 0.587×G + 0.114×B. We apply BGR to grayscale on our dataset using the code below: gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY) We then apply Gaussian blur. Gaussian blur is a popular image processing technique for lowering noise and detail.A smooth appearance is produced by averaging pixel values using a Gaussian function. This method effectively lowers high-frequency noise and softens backgrounds. The Gaussian Blur method is applied to our dataset using the code below: gray = cv2.GaussianBlur(gray, (5,5), 0) Additionally, we established a threshold value as indicated below; thres = cv2.threshold(gray, 45, 255, cv2.THRESH_BINARY) It cuts off a certain intensity level from grayscale photos to produce binary pictures. Pixels with values below the threshold are altered to a different color (often black), while those with values above the threshold are changed to one color (typically white). The image processing phase involves several key steps to enhance analysis. Object detection is the initial step, which focuses on isolating relevant objects from the background to facilitate targeted examination. This is followed by image segmentation , where the image is divided into distinct regions, allowing for more precise analysis of individual components. Next, feature extraction simplifies the image by emphasizing specific characteristics, making important features easier to identify and interpret. Finally, noise reduction is applied to eliminate minor variations and enhance the clarity of significant structures within the image, ensuring more accurate results in subsequent processing stages. IV. MODEL ARCHITECTURE SELECTION In our model, VGG-19 is utilized. VGG-19's deep layers and excellent visual classification make it a powerful CNN architecture. In the design, fully linked layers come after convolutional and pooling layers. The CNN architecture, which has three output layers and sixteen input layers, will be built using the VGG-19 model. This VGG-19 model will be trained using a predefined training set to provide a training model that will be used for future lung cancer diagnosis. A. Model Training This level consists of 4 steps. The CNN's training process, which makes use of the MMT-Vin model based on VGG-19, will be briefly reviewed in this article. B. Input To ensure a constant input size, our model can handle images of 224 x 224 pixel sizes. C. Convolution Layers The VGG-19-based MMT-Vin basic has 19 layers, including convolutional, max-pooling, and fully linked layers, as the name suggests. In order to extract visual features, convolutional layers are necessary. Blocks with numerous convolutional and max-pooling layers make up the network. Convolutional layers employ 3x3 filters with a stride of 1 and no padding in order to maintain the input spatial dimensions.Each convolutional block adds layers as we move through the network, enabling the model to pick up increasingly complex traits. VGG-19 consists of 16 convolutional layers and 5 ConvBlocks. D. Max-Pooling layers Following each convolutional layer, VGG-19 employs max-pooling layers with a window size of 2x2 and a stride of two. These max-pooling layers reduce processing costs while capturing translation-invariant information by downsampling the spatial dimensions of feature maps. E. Fully-connected layers Every layer has neurons that are highly connected to every other layer's neurons. The number of neurons in fully connected layers declines as categorization advances in the direction of the output layer, which correlates with output classification or categories. F. Model Evaluation The image dataset will be used to evaluate performance at every stage. To calculate recall, accuracy, precision, and loss, the testing dataset will be input into the trained CNN model. G. Deployment The model can recognize pictures of lung cancer when it functions properly. Histopathology images are utilized to detect lung cancer during deployment. The CNN, a deep learning model, receives the unique features of the input image. This stage classifies images of lung cancer using the CNN model. V. RESULTS AND DISCUSSION For the experiment in this study, pictures of both benign and malignant lung cancer were gathered. The dataset needs to be diverse and well-labeled for the CNN model to learn well. Kaggle was used to obtain images of lung cancer histology for this investigation. The gathering process involves 4250 pre-processed and tagged lung cancer photos. In order to streamline the training and testing procedure, 70 percent of the dataset will be used for training, 15percent for testing, and 15percent for validation.The accuracy, loss, precision, and recall of the CNN model will be computed using the testing dataset. This process evaluates the CNN model's performance. This section compares the lung cancer prediction model's performance across two experimental sets with different epoch values. The study emphasizes the model's convergence behavior in relation to the number of training epochs. The model's performance reaches a plateau and more training repetitions yield declining results after the training process reaches a point of convergence. The convergence patterns seen with 20 and 25 epochs can be compared to identify the optimal training length for achieving satisfactory performance. Specific performance scores for 20 epochs are given in Table 2 , while performance scores for 25 epochs are shown in Table 3 . Table 2 Results of the proposed CNN's performance using VGG-19 (epoch 20) Accuracy Precision Loss Recall 0.9342 0.92 0.16 0.92 Table 3 Results of the proposed CNN's performance using VGG-19 (epoch 25) Accuracy Precision Loss Recall 0.9758 0.97 0.12 0.97 As a result, the findings provide an overview of the trials conducted over two distinct time periods (20 and 25 epochs) and emphasise their significance in raising the lung cancer prediction model's accuracy in comparison to earlier research. VI. RESULT COMPARISON WITH NORMAL VGG19 A collection of 15,000 histological images—10,000 of which were of malignant lung cancer and 5,000 of which were of benign tissue—were used to train and evaluate the classifier. The experimental results show that the VGG-19 architecture can achieve an accuracy rate of 95% [ 30 ]. Whereas our model uses over 4000 images of which approx. 2000 were benign and approx. 2000 were of malignant. By experimental findings our model can get 97.58% accuracy. VII. CONCLUSION Deep learning's ability to independently build features is by far its greatest advantage over other machine learning methods. Because of these characteristics, it can analyse data to identify similar properties and incorporate them for faster learning. CNNs and other deep learning techniques have transformed medical image processing in recent years. CNNs are especially well-suited for image-based medical diagnostics because of their capacity to decipher intricate visual patterns and extract valuable information from pictures. The CNN algorithm will categorise the input lung image as normal or abnormal after a successful training and testing phase. Lung cancer from histopathological scans was predicted using a Deep Convolutional Neural Network (CNN) built on the VGG-19 model. Rather than being a separate model from CNNs, VGG-19 is a specific architecture within the CNN framework. When evaluating CNN designs, performance is a crucial consideration. The experiment's findings show that the CNN with VGG-19 performs better than other CNNs. Future studies should concentrate on adding more datasets of lung cancer images and developing a more accurate model for predicting lung cancer from these images. Declarations Here is the data set requested. https://www.kaggle.com/datasets/chinmay2168/lung-cancer-images Author Contribution Vinayak: Conceptualization, Data curation, Methodology, Software implementation, Formal analysis, Visualization, Writing – original draft.Manish Madhava Tripathi: Supervision, Validation, Writing – review & editing, Project administration, Resources. References K. Zou et al., “National Cancer Center / National Clinical Research Center for Cancer / Cancer Yale School of Public Health, New Haven, USA,” pp. 0–37, 2022, doi: 10.1016/j.jncc.2022.09.004 . P. R. Radhika, R. A. S. Nair, and G. Veena, “A Comparative Study of Lung Cancer Detection using Machine Learning Algorithms,” Proceedings of 2019 3rd IEEE International Conference on Electrical, Computer and Communication Technologies, ICECCT 2019, pp. 1–4, 2019, doi: 10.1109/ICECCT.2019.8869001 . Hadrien T. Gayap and Moulay A. Akhloufi, “Deep Machine Learning for Medical Diagnosis, Application to Lung Cancer Detection: A Review”, BioMedInformatics Jan 2024, https://doi.org/10.3390/biomedinformatics4010015 . S. Chen, Y. Han, J. Lin, X. Zhao, and P. Kong, “Pulmonary nodule detection on chest radiographs using balanced convolutional neural network and classic candidate detection,” Artificial Intelligence in Medicine, vol. 107, no. May, p. 101881, 2020, doi: 10.1016/j.artmed.2020.101881 . Y. Gu et al., “A survey of computer-aided diagnosis of lung nodules from CT scans using deep learning,” Computers in Biology and Medicine, vol. 137, no. March, 2021, doi: 10.1016/j.compbiomed.2021.104806 . S. Pang and X. U. N. Wang, “A Deep Model for Lung Cancer Type Identification by Densely Connected Convolutional Networks and Adaptive Boosting,” IEEE Access, vol. 8, pp. 4799–4805, 2020, doi: 10.1109/ACCESS.2019.2962862 . J. Feng, A. W. Godana, S. Liu, and D. J. Gelmecha, “DFD-Net : lung cancer detection from denoised CT scan image using deep learning,” vol. 15, no. 2, 2021. O. Akter, M. A. Moni, M. M. Islam, J. M. W. Quinn, and A. H. M. Kamal, “Lung cancer detection using enhanced segmentation accuracy,” no. October, 2020. S. Pang et al., “VGG16-T : A Novel Deep Convolutional Neural Network with Boosting to Identify Pathological Type of Lung Cancer in Early Stage by CT Images,” vol. 13, no. 1, pp. 771–780, 2020. Tripathi, M. M., Haroon, M., Khan, Z., & Husain, M. S. (2022). Security in digital healthcare system. Pervasive Healthcare: A Compendium of Critical Factors for Success, 217–231. Manish Madhava Tripathi, Saurabh Pandey, “Diagnosis of Diabetes using Artificial Intelligence Techniques by using Bio Medical Signal Data”, International Journal of Research and Development in Applied Science and Engineering (IJRDASE) ISSN-2454-6844, Volume 13, Issue 2, May 2017 Liangyu Lia, Jing Yang et. al.(2024), “Enhancing lung cancer detection through hybrid features and machine learning hyperparameters optimization techniques”, https://doi.org/10.1016/j.heliyon.2024.e26192 J. Redmon and A. Farhadi, “YOLO9000: Better, faster, stronger,” Proceedings – 30th IEEE Conference on Computer Vision and Pattern Recognition, CVPR 2017, vol. 2017-Janua, pp. 6517–6525, 2017, doi: 10.1109/CVPR.2017.690 . M. Toğaçar, “Disease type detection in lung and colon cancer images using the complement approach of inefficient sets,” Computers in Biology and Medicine, vol. 137, p. 104827, 2021, doi: 10.1016/j.compbiomed.2021.104827 . R. Grossman, O. Haim, S. Abramov, and B. Shofty, “Differentiating Small-Cell Lung Cancer From Non-Small-Cell Lung Cancer Brain Metastases Based on MRI Using Efficientnet and Transfer Learning Approach,” vol. 20, pp. 1–7, 2021, doi: 10.1177/15330338211004919 . H. Yang et al., “Deep learning-based six-type classifier for lung cancer and mimics from histopathological whole slide images : a retrospective study,” pp. 1–14, 2021. K. Muhammad, H. Ullah, A. Khader, and J. Saudagar, “WEENet : An Intelligent System for Diagnosing COVID-19 and Lung Cancer in IoMT Environments,” vol. 11, no. February, pp. 1–13, 2022, doi: 10.3389/fonc.2021.811355 . A. Lang et al., “Informatics in Medicine Unlocked Model architecture and tile size selection for convolutional neural network training for non-small cell lung cancer detection on whole slide images,” Informatics in Medicine Unlocked, vol. 28, p. 100850, 2022, doi: 10.1016/j.imu.2022.100850 . L. Techniques, “A CAD System for Lung Cancer Detection Using Hybrid Deep,” 2023. A. Elnakib, H. M. Amer, and F. E. Z. Abou-chadi, “Early Lung Cancer Detection Using Deep Learning Optimization,” pp. 82–94. M. Humayun, R. Sujatha, S. N. Almuayqil, and N. Z. Jhanjhi, “A Transfer Learning Approach with a Convolutional Neural Network for the Classification of Lung Carcinoma,” pp. 1–15, 2022. G. M. M. Alshmrani, Q. Ni, R. Jiang, H. Pervaiz, and N. M. Elshennawy, “A deep learning architecture for multi-class lung diseases classification using chest X-ray (CXR) images,” Alexandria Engineering Journal, vol. 64, pp. 923–935, 2023, doi: 10.1016/j.aej.2022.10.053 . Jiang, Y., Ebrahimpour, L., Després, P. et al. (2025). A benchmark of deep learning approaches to predict lung cancer risk using national lung screening trial cohort. Sci Rep 15, 1736. https://doi.org/10.1038/s41598-024-84193-7 Li, L., Yang, J., Por, L. Y., Khan, M. S., Hamdaoui, R., Hussain, L., Iqbal, Z., Rotaru, I. M., Dobrotă, D., Aldrdery, M., & Omar, A. (2024). Enhancing lung cancer detection through hybrid features and machine learning hyperparameters optimization techniques. Heliyon, 10(4). https://doi.org/10.1016/j.heliyon.2024.e26192 Linh, V. T. N., Kim, H., Lee, M. Y., Mun, J., Kim, Y., Jeong, B. H., Park, S. G., Kim, D. H., Rho, J., & Jung, H. S. (2024). 3D plasmonic hexaplex paper sensor for label-free human saliva sensing and machine learning-assisted early-stage lung cancer screening. Biosensors and Bioelectronics, 244(August 2023), 115779. https://doi.org/10.1016/j.bios.2023.115779 Gayap, H. T., & Akhloufi, M. A. (2024). Deep Machine Learning for Medical Diagnosis, Application to Lung Cancer Detection: A Review. BioMedInformatics, 4(1), 236–284. https://doi.org/10.3390/biomedinformatics4010015 Javed, R., Abbas, T., Khan, A. H., Daud, A., Bukhari, A., & Alharbey, R. (2024). Deep learning for lungs cancer detection: a review. In Artificial Intelligence Review (Vol. 57, Issue 8). Springer Netherlands. https://doi.org/10.1007/s10462-024-10807-1 Ashraf, S. F., Yin, K., Meng, C. X., Wang, Q., Wang, Q., Pu, J., & Dhupar, R. (2022). Predicting benign, preinvasive, and invasive lung nodules on computed tomography scans using machine learning. Journal of Thoracic and Cardiovascular Surgery, 163(4), 1496–1505.e10. https://doi.org/10.1016/j.jtcvs.2021.02.010 Gopinath, A., Gowthaman, P., Venkatachalam, M., & Saroja, M. (2023). Computer aided model for lung cancer classification using cat optimized convolutional neural networks. Measurement: Sensors, 30(January), 100932. https://doi.org/10.1016/j.measen.2023.100932 Saranya, N., Kanthimathi, N., Boomika, S., Bavatharani, S., Karthick raja, R. (2022). Classification and Prediction of Lung Cancer with Histopathological Images Using VGG-19 Architecture. In: Kalinathan, L., R., P., Kanmani, M., S., M. (eds) Computational Intelligence in Data Science. ICCIDS 2022. IFIP Advances in Information and Communication Technology, vol 654. Springer, Cham. https://doi.org/10.1007/978-3-031-16364-7_12 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-7488263","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":512704807,"identity":"d231981e-23d9-44ac-ab5b-b45810e65bdd","order_by":0,"name":"Vinayak Vinayak","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA9UlEQVRIiWNgGAWjYLCCBBDBw3yMmYGNgQcqJkGMFrY0sBYeorSAAQ+PGUgL3BqcQH726TSJhzvsEvt7znx7XFB2R8Ze7ADjhx8MFnm4tBicy90mkXgmOXHG2d7txjPOPePhkU5gluxhkCjGqYWHd7NBYhtzYsN53m3SvG2HQVoYpIF+SWzA5bAesJb6xPnneZ7BtDD/xqeF4QzvxgeJbYcTN5ztYYNpYcNriwFEy3HjjWeOmQP9AtRyO7HNsscAr8M2HPzZVi0770zyM2CIHbZnn518+MaPijrcDoMCRyQFjEC2AQH1QGBPWMkoGAWjYBSMWAAA/nJUY24lyUYAAAAASUVORK5CYII=","orcid":"","institution":"Dr. A.P.J. Abdul Kalam Technical University","correspondingAuthor":true,"prefix":"","firstName":"Vinayak","middleName":"","lastName":"Vinayak","suffix":""}],"badges":[],"createdAt":"2025-08-29 11:38:18","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-7488263/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7488263/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":91200622,"identity":"cc494419-e299-49b1-80e4-93ad077b0c8f","added_by":"auto","created_at":"2025-09-12 15:29:02","extension":"jpg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":27681,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eStages in the Developed MMT-Vin Basic Framework\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"Picture1.jpg","url":"https://assets-eu.researchsquare.com/files/rs-7488263/v1/7af0d2a405e38741c1940091.jpg"},{"id":91198088,"identity":"fce2a0e2-0cb8-49a0-886e-c5355005d644","added_by":"auto","created_at":"2025-09-12 15:13:02","extension":"jpg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":83134,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eSuggested MMT-Vin Basic Framework\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"Picture2.jpg","url":"https://assets-eu.researchsquare.com/files/rs-7488263/v1/1f9ec4662f5099aa8f0ed635.jpg"},{"id":91198091,"identity":"185fa445-4635-49de-8ca2-2feb4660604c","added_by":"auto","created_at":"2025-09-12 15:13:02","extension":"jpg","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":18797,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eFlow diagram of Original VGG19\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"Picture3.jpg","url":"https://assets-eu.researchsquare.com/files/rs-7488263/v1/903415dbd9e6d5178825ccb6.jpg"},{"id":91200623,"identity":"4ae939cb-714a-40b4-8998-a75ba0526f1b","added_by":"auto","created_at":"2025-09-12 15:29:02","extension":"jpg","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":25748,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eGraph plot of CT scan images before and after augmentation\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"Picture4.jpg","url":"https://assets-eu.researchsquare.com/files/rs-7488263/v1/e50bd7ffcacec7042dcc80a1.jpg"},{"id":91199600,"identity":"8c8cc80d-f316-4df9-b78e-e06c64cb4ab0","added_by":"auto","created_at":"2025-09-12 15:21:02","extension":"jpg","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":33111,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eUnprocessed and cropped versions of a lung CT scan\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"Picture5.jpg","url":"https://assets-eu.researchsquare.com/files/rs-7488263/v1/30a78f3c1b2d835556cfc75b.jpg"},{"id":91198095,"identity":"37447b87-4d85-4f06-aed3-2dab5e150996","added_by":"auto","created_at":"2025-09-12 15:13:02","extension":"jpg","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":59536,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eTraining Accuracy and Loss Gap\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"Picture6.jpg","url":"https://assets-eu.researchsquare.com/files/rs-7488263/v1/3fb5b2780a6ccce6b864dba0.jpg"},{"id":92393306,"identity":"b7718bbb-1a31-4b66-afdb-c944d4039832","added_by":"auto","created_at":"2025-09-29 08:59:23","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":812076,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7488263/v1/55687cc0-a506-4aac-a44d-abca19f1ca03.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"A Data-Driven Approach to Detecting Lung Cancer with Smart VGG19 Algorithm","fulltext":[{"header":"I. INTRODUCTION","content":"\u003cp\u003eThe primary cause of cancer-related death worldwide is still lung cancer. Risk factors for lung cancer include radiation therapy, alcohol, radon gas, asbestos, secondhand smoke, smoking, radiation therapy, and a family history of the disease [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]. Survival chances for lung cancer can be considerably increased by early detection. Image processing, machine learning, or hybrid approaches from the literature are used by researchers to detect lung cancer by Naive Bayes (NB), Convolutional Neural Networks (CNN), Support Vector Machines (SVM), Decision Trees (DT), Back Propagation Networks (BPN), Logistic Regression (LR), and K-Nearest Neighbors (KNN) [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]. Compared to other neural networks, CNNs have fewer parameters thanks to their pooling layers and local receptive fields. They decrease false positives and do away with manual nodule signifiers [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eChest radiography (CXR), the most widely used imaging diagnostic method, can be used to diagnose disorders of the chest, including emphysema, pneumonia, pneumoconiosis, lung cancer, and tuberculosis. It is the most widely used, least expensive, and least radioactive diagnostic method that can identify undiscovered disease alterations [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e].Computed tomography (CT) has a high sensitivity when compared to computed radiography (CR), and early CT screening has a remarkable influence on the diagnosis of lung cancer [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]. In [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e], a deep learning method for predicting the kind of lung cancer from CT scans was presented. This model classifies malignant tumors using densely linked convolutional networks (DenseNet), and it uses adaptive boosting (adaboost) to aggregate classification results for better performance. With 89.95% accuracy, the DenseNet model beat the VGG16, AlexNet, ResNet, and DenseNet without a boost.\u003c/p\u003e\u003cp\u003eIn another study, the authors of [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e] presented a model with a denoising and detection component termed DFD-Net (\"denoising first\" two-path convolutional neural network). The DR-Net model, a residual learning denoising model, is used to initially remove the noise during the pre-processing phase. To detect lung cancer, the denoised images from the DR-Net model are subsequently fed into a two-path convolutional neural network. Discriminant correlation analysis enhances this model's performance by include more representative features, as opposed to the conventional feature concatenation method, which concatenates two sets of data from different CNN layers. Lastly, a retraining approach is suggested to get rid of problems related to the picture label imbalance. An improved CT scan method for lung nodule detection of CAD was presented in another study [\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e]. This method segments lung nodules from CT scans using a fuzzy-based algorithm. Fuzzy soft membership functions are used to automate segmentation. Lung nodules are reliably classified as benign or malignant by fuzzy neural classifiers. [\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e] Presented a poor classifier, and multiple VGG16-T neural networks are trained using the boosting technique. Accuracy is increased by combining weak classifier VGG16-T networks.\u003c/p\u003e"},{"header":"II. Literature Review","content":"\u003cp\u003eBecause machine learning algorithms make it possible to analyze medical data more quickly and accurately, they are revolutionizing healthcare diagnostics. These algorithms can help clinicians make well-informed judgments by spotting trends and anomalies in big datasets [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]. Predictive analytics for patient outcomes, individualized treatment strategies, and early disease identification are some examples of applications. Healthcare systems can increase the efficiency of patient care and improve diagnostic accuracy by utilizing machine learning [\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e]. Researchers have used machine learning and deep learning for categorisation and prediction across imaging and signal modalities. Machine learning algorithms are being used more and more in medical diagnostic systems to forecast lung conditions [\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eDarkNet-19 is a deep learning network that uses YOLO object detection [\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e]. There are five pooling layers and 19 convolutional layers in DarkNet-19. Similar to VGG models, the DarkNet-19 model has twice as many channels but employs 3x3 filters following convolutional layers. Batch normalization is another technique used by the DarkNet-19 model to stabilize training, accelerate convergence, and simplify model operation. In another work, the DarkNet-19 model, a deep learning model, was used to train the image classes from scratch [\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]. Using the feature set obtained from the DarkNet-19 model, the unsuccessful features were selected using the optimisation techniques of Manta Ray Foraging and Equilibrium.\u003c/p\u003e\u003cp\u003eComplementary rule insets are feature sets that remain efficient after the inefficient features have been isolated from the rest of the set.The two employed the following successful characteristics: integrating optimization methods with the SVM classification approach.\u003c/p\u003e\u003cp\u003eThe EfficientNet model, which uses a technique known as compound scaling to modify all dimensions, is examined by the authors of [\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e]. In this strategy, FLOPs and accuracy are optimized by a multi-objective architectural search. EfficientNet employs a search space and the ACC (m) multiplyby [FLOPS (m)/T]w metric for optimization. FLOPS and ACC and symbolize the accuracy and FLOPs of model \"m,\" while T and wstand for the preferred hyper-parameters and FLOPs. These elements are essential for striking a balance between computing efficiency and accuracy. This network consists of several convolutional layers with varying kernel sizes.\u003c/p\u003e\u003cp\u003eThe first input frame is 224 \u0026times; 224 \u0026times; 3 and has three color channels (red, green, and blue). As layers develop, width increases to boost accuracy and resolution falls to condense feature maps. This method gathers important input frame characteristics. Kernels are 112 pixels wide in 2nd layer and 64 pixels wide in the subsequent convolutional layer. To capture the most noticeable features, the final layer employs up to 2,560 kernels at a decreased resolution of 7 \u0026times; 7. For classification, SoftMax layers, encoding, and max pooling are added.\u003c/p\u003e\u003cp\u003eGoogle unveiled the EfficientNet [\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e], a collection of convolutional neural networks (CNNs) developed using an automated neural network architecture search. CNNs are designed more efficiently because to this automated method, which produces models with improved performance and lower processing costs. All of the models in the EfficientNet family (EfficientNet b0 through b7) are designed to operate best with particular image dimensions in order to balance efficiency and performance. For this study, EfficientNet b3 was chosen since its input size is similar to that of the InceptionResNetV2 model. According to [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e], lung nodules in low-dose computed tomography (LDCT) images can be found early by employing computer-aided detection (CADe). To improve the contrast of low-dose images, our proposed method starts with data processing. Analysing deep learning architectures such as VGG16, VGG19, and Alex produces succinct features.\u003c/p\u003e\u003cp\u003eOn the dataset, all transfer learning techniques\u0026mdash;including VGG 16, VGG 19, and Xception with a 20-epoch structure\u0026mdash;perform admirably. Preprocessing facilitates the creation of a reliable model and speeds up model integration, allowing for the prediction of future events. The accuracy of VGG 16 is 98.83 percent by the twentieth epoch, 98.05 percent by VGG 19, and 97.4 percent by Xception. GoramMufarah M. Alshmrani et al. [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e] employed a fully connected network for classification, three CNN blocks for feature extraction, and a pre-trained VGG19 model.\u003c/p\u003e\u003cp\u003eThe use of deep learning (DL) algorithms to analyse CT scan data has lately proved essential in enhancing lung cancer detection, especially in the categorisation of malignancy. Deep learning methods in medical imaging fall into two main categories: 2D and 3D models. The channel is the colour channel number, which is typically one for greyscale radiological images and three for natural photos in RGB scale. Two-dimensional deep learning models use two-dimensional images in a (channel, width, height) format [\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eIn terms of lung cancer detection, machine learning holds enormous potential for enabling early identification and potentially improving patient outcomes. In this area, feature extraction remains a major challenge. Combining the most relevant features can increase detection accuracy even more. This work employs a hybrid feature extraction method that combines Gray-level co-occurrence matrix (GLCM) with Haralick and autoencoder features. Afterwards, supervised machine learning methods were trained using these attributes as input. GLCM with autoencoder, Haralick, and autoencoder features yielded an accuracy of 99.89% for SVM polynomial, but flawless performance metrics were achieved by SVM Gaussian and SVM Radial Base Function (RBF) [\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eA label-free detection approach for noninvasive biofluids enables early-stage cancer diagnosis and rapid on-site disease screening by looking at metabolic alterations. Here, we develop three-dimensional plasmonic hexaplex nanostructures that are deposited onto a 3D-PHP paper substrate. This adaptable and extremely absorbent 3D-PHP sensor is used in conjunction with a commercial saliva collection tube to create an efficient on-site sensing platform for lung cancer screening by detecting human saliva using surface-enhanced Raman scattering (SERS). The enhanced interface between the multispikehexaplex-shaped gold nanostructure and saliva viscosity enables efficient sampling and SERS enhancement. Through the analysis of patient salivary samples, the 3D-PHP sensor effectively identifies and diagnoses lung cancer. A logistic regression-based machine learning algorithm successfully differentiates between benign and malignant patients, demonstrating high clinical sensitivity and specificity [\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eDeep learning has emerged as a powerful tool for medical image analysis and diagnosis, according to the author of [\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e], who also noted that deep learning performs well on tasks. This literature review summarises recent studies on deep learning techniques for lung cancer screening and diagnosis. Significant advancements, limitations, and potential future directions in the state-of-the-art in deep learning for lung cancer diagnosis are highlighted in this work. We prioritised studies that provided a comprehensive view of the area by utilising significant public datasets such as LUNA16, JSRT, and LIDC. We focus on deep learning architectures which include 2D and 3D convolutional neural networks (CNNs), vision transformers (ViT), dual-path networks, and natural language processing (NLP).\u003c/p\u003e\u003cp\u003eA comprehensive Systematic Literature Review (SLR) for lung cancer research is conducted by this study [\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e] using deep learning techniques. It provides a detailed analysis of the methodology, cutting-edge developments, quality assessments, and customised deep learning approaches. While there are many deep learning methods for classifying lung cancer, this study focusses mostly on the most popular method, the Convolutional Neural Network (CNN). The highest accuracy is achieved by CNN due to its multi-layer architecture, automated weight learning, and capacity to transfer local weights. CNN consistently shows the highest accuracy when performance parameters like precision, accuracy, specificity, sensitivity, and AUC are shown for multiple methods. The findings show that DCNN boosts lung cancer detection and classification greatly.\u003c/p\u003e\u003cp\u003eThe models that classified nodules into six groups using convolutional neural networks and support vector machines had an area under the curve of 0.89/0.87 between atypical adenomatous hyperplasia\u0026thinsp;+\u0026thinsp;adenocarcinoma in situ and minimally invasive adenocarcinoma and invasive adenocarcinoma, 0.87/0.86 between minimally invasive adenocarcinoma and invasive adenocarcinoma, 0.76/0.72 between atypical adenomatous hyperplasia\u0026thinsp;+\u0026thinsp;adenocarcinoma in situ and minimally invasive adenocarcinoma, and 0.93/0.92 between atypical adenomatous hyperplasia\u0026thinsp;+\u0026thinsp;adenocarcinoma in situ and minimally invasive adenocarcinoma. For the categorisation of adenocarcinoma in situ, minimally invasive versus invasive cancer, and benign versus atypical adenomatous hyperplasia, the support vector machine/convolutional neural network models yielded a micro-average area under the curve of 0.93/0.94 [\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eAdditional useful information about a lung cancer diagnosis can be obtained from images from computer tomography (CT) scans. Using CT scan input images, several Machine Learning (ML) and Deep Learning (DL) algorithms are created to enhance diagnosis and treatment procedures. But the hardest aspect of research is still creating a sophisticated and accurate system. This paper proposes a novel classification technique called Deep Fused Features-Based Cat-Optimized Networks (DFF-CON), which is based on the concepts of fused features and optimised networks. In the proposed structure, Deep Convolutional Neural Networks (DCNN) are employed to enhance the classification maps and ultimately decrease the probability of an overfitting issue [\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e].\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eAnalysis and Comparison of Different CNN Models\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"4\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eAuthor Name\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eApproach\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eAdvantages\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eLimitations\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eJ. Redmonet. al. [\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e], M. Toğa\u0026ccedil;ar[\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eDeep learning using DarkNet-19 with YOLO object detection and feature optimization (MRFO\u0026thinsp;+\u0026thinsp;Equilibrium)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eEfficient feature extraction, faster convergence, improved training stability\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eComputational complexity; requires deep learning expertise\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eK. Muhammadet. al.[\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e], A. Langet. al.[\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eCompound scaling in CNNs; multi-objective architecture search for model optimization\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eHigh accuracy with low computational cost; scalable across devices\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eArchitecture complexity; requires careful tuning\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eA. Elnakibet. al.[\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eCADe system using VGG16, VGG19, and AlexNet on LDCT scans\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eEarly-stage detection, improved image contrast through preprocessing\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eRelatively shallow networks compared to newer models\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eM. Humayunet. al.[\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eNon-invasive diagnostic strategy with transfer learning (VGG16, VGG19, Xception)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eFewer parameters, robust on varying dataset sizes, suitable for clinical evaluation\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eMay still require high-end computation for training\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eG. M. M. Alshmraniet. al. [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eThree CNN blocks for feature extraction, a fully linked network for classification, and a pre-trained VGG19\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eEffective feature learning, structured model pipeline\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eRequires high-quality pre-training and labeled datasets\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003eThis work aims to create a trustworthy algorithm for the diagnosis of lung cancer that provides improved sensitivity and specificity while retaining low time complexity, motivated by current research initiatives. The work's specifics and the suggested methodology are described in the following sections.\u003c/p\u003e"},{"header":"III. PROPOSED METHODOLOGY","content":"\u003cp\u003eThe suggested method for detecting lung cancer is divided into several stages, each of which is described in turn. An overview of the process workflow is introduced initially.After that data set collection and data preprocessing is discussed. Model architecture selection and model training explained later. Model evaluation, deployment discussed next. At the end results and discussion and conclusion given in this paper.\u003c/p\u003e\u003cp\u003e\u003cem\u003eA. Overview of the Process Workflow\u003c/em\u003e\u003c/p\u003e\u003c/li\u003e\u003c/ul\u003e\u003c/p\u003e\u003cp\u003eClassifying lung cancer using a CNN based on the MMT-Vin model with VGG-19 architecture involves several stages, including dataset collection, image preprocessing, model training, and others. Figures\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e and \u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e provide a concise overview of the proposed method.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003e\u003cem\u003eB. Dataset Collection\u003c/em\u003e\u003c/p\u003e\u003cp\u003eA large collection of lung cancer photos, including both benign and malignant examples, is shown in this section. Accurate labeling and diversity in the dataset are critical for the CNN model to learn efficiently. During the data collection stage of this study, the lung cancer histology photos were taken from Kaggle. Upon unzipping, we discovered that there are 515 pictures of lung cancer in the collection. After augmentation, we were able to locate 4,250 images.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003e\u003cem\u003eC. Data Pre-processing\u003c/em\u003e\u003c/p\u003e\u003cp\u003eFirst, we convert BGR (blue, green, and red) photos to grayscale. To generate a single intensity value for every pixel, the conversion employs distinct weights for every channel. The often used formula is\u003c/p\u003e\u003cp\u003egray\u0026thinsp;=\u0026thinsp;0.299\u0026times;R\u0026thinsp;+\u0026thinsp;0.587\u0026times;G\u0026thinsp;+\u0026thinsp;0.114\u0026times;B.\u003c/p\u003e\u003cp\u003eWe apply BGR to grayscale on our dataset using the code below:\u003c/p\u003e\u003cp\u003egray\u0026thinsp;=\u0026thinsp;cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)\u003c/p\u003e\u003cp\u003eWe then apply Gaussian blur. Gaussian blur is a popular image processing technique for lowering noise and detail.A smooth appearance is produced by averaging pixel values using a Gaussian function. This method effectively lowers high-frequency noise and softens backgrounds.\u003c/p\u003e\u003cp\u003eThe Gaussian Blur method is applied to our dataset using the code below:\u003c/p\u003e\u003cp\u003egray\u0026thinsp;=\u0026thinsp;cv2.GaussianBlur(gray, (5,5), 0)\u003c/p\u003e\u003cp\u003eAdditionally, we established a threshold value as indicated below;\u003c/p\u003e\u003cp\u003e\u003cb\u003ethres\u0026thinsp;=\u0026thinsp;cv2.threshold(gray, 45, 255, cv2.THRESH_BINARY)\u003c/b\u003e\u003c/p\u003e\u003cp\u003eIt cuts off a certain intensity level from grayscale photos to produce binary pictures. Pixels with values below the threshold are altered to a different color (often black), while those with values above the threshold are changed to one color (typically white).\u003c/p\u003e\u003cp\u003eThe image processing phase involves several key steps to enhance analysis. \u003cb\u003eObject detection\u003c/b\u003e is the initial step, which focuses on isolating relevant objects from the background to facilitate targeted examination. This is followed by \u003cb\u003eimage segmentation\u003c/b\u003e, where the image is divided into distinct regions, allowing for more precise analysis of individual components. Next, \u003cb\u003efeature extraction\u003c/b\u003e simplifies the image by emphasizing specific characteristics, making important features easier to identify and interpret. Finally, \u003cb\u003enoise reduction\u003c/b\u003e is applied to eliminate minor variations and enhance the clarity of significant structures within the image, ensuring more accurate results in subsequent processing stages.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e"},{"header":"IV. MODEL ARCHITECTURE SELECTION","content":"\u003cp\u003eIn our model, VGG-19 is utilized. VGG-19's deep layers and excellent visual classification make it a powerful CNN architecture. In the design, fully linked layers come after convolutional and pooling layers. The CNN architecture, which has three output layers and sixteen input layers, will be built using the VGG-19 model. This VGG-19 model will be trained using a predefined training set to provide a training model that will be used for future lung cancer diagnosis.\u003c/p\u003e\u003cp\u003e\u003cem\u003eA. Model Training\u003c/em\u003e\u003c/p\u003e\u003cp\u003eThis level consists of 4 steps. The CNN's training process, which makes use of the MMT-Vin model based on VGG-19, will be briefly reviewed in this article.\u003c/p\u003e\u003cp\u003e\u003cem\u003eB. Input\u003c/em\u003e\u003c/p\u003e\u003cp\u003eTo ensure a constant input size, our model can handle images of 224 x 224 pixel sizes.\u003c/p\u003e\u003cp\u003e\u003cem\u003eC. Convolution Layers\u003c/em\u003e\u003c/p\u003e\u003cp\u003eThe VGG-19-based MMT-Vin basic has 19 layers, including convolutional, max-pooling, and fully linked layers, as the name suggests. In order to extract visual features, convolutional layers are necessary. Blocks with numerous convolutional and max-pooling layers make up the network. Convolutional layers employ 3x3 filters with a stride of 1 and no padding in order to maintain the input spatial dimensions.Each convolutional block adds layers as we move through the network, enabling the model to pick up increasingly complex traits. VGG-19 consists of 16 convolutional layers and 5 ConvBlocks.\u003c/p\u003e\u003cp\u003e\u003cem\u003eD. Max-Pooling layers\u003c/em\u003e\u003c/p\u003e\u003cp\u003eFollowing each convolutional layer, VGG-19 employs max-pooling layers with a window size of 2x2 and a stride of two. These max-pooling layers reduce processing costs while capturing translation-invariant information by downsampling the spatial dimensions of feature maps.\u003c/p\u003e\u003cp\u003e\u003cem\u003eE. Fully-connected layers\u003c/em\u003e\u003c/p\u003e\u003cp\u003eEvery layer has neurons that are highly connected to every other layer's neurons. The number of neurons in fully connected layers declines as categorization advances in the direction of the output layer, which correlates with output classification or categories.\u003c/p\u003e\u003cp\u003e\u003cem\u003eF. Model Evaluation\u003c/em\u003e\u003c/p\u003e\u003cp\u003eThe image dataset will be used to evaluate performance at every stage. To calculate recall, accuracy, precision, and loss, the testing dataset will be input into the trained CNN model.\u003c/p\u003e\u003cp\u003e\u003cem\u003eG. Deployment\u003c/em\u003e\u003c/p\u003e\u003cp\u003eThe model can recognize pictures of lung cancer when it functions properly. Histopathology images are utilized to detect lung cancer during deployment. The CNN, a deep learning model, receives the unique features of the input image. This stage classifies images of lung cancer using the CNN model.\u003c/p\u003e"},{"header":"V. RESULTS AND DISCUSSION","content":"\u003cp\u003eFor the experiment in this study, pictures of both benign and malignant lung cancer were gathered. The dataset needs to be diverse and well-labeled for the CNN model to learn well. Kaggle was used to obtain images of lung cancer histology for this investigation. The gathering process involves 4250 pre-processed and tagged lung cancer photos. In order to streamline the training and testing procedure, 70 percent of the dataset will be used for training, 15percent for testing, and 15percent for validation.The accuracy, loss, precision, and recall of the CNN model will be computed using the testing dataset. This process evaluates the CNN model's performance.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003eThis section compares the lung cancer prediction model's performance across two experimental sets with different epoch values. The study emphasizes the model's convergence behavior in relation to the number of training epochs. The model's performance reaches a plateau and more training repetitions yield declining results after the training process reaches a point of convergence. The convergence patterns seen with 20 and 25 epochs can be compared to identify the optimal training length for achieving satisfactory performance. Specific performance scores for 20 epochs are given in Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e, while performance scores for 25 epochs are shown in Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e.\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eResults of the proposed CNN's performance using VGG-19 (epoch 20)\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"4\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eAccuracy\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003ePrecision\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eLoss\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eRecall\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e0.9342\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e0.92\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.16\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.92\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab3\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eResults of the proposed CNN's performance using VGG-19 (epoch 25)\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"4\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eAccuracy\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003ePrecision\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eLoss\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eRecall\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e0.9758\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e0.97\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.12\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.97\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003eAs a result, the findings provide an overview of the trials conducted over two distinct time periods (20 and 25 epochs) and emphasise their significance in raising the lung cancer prediction model's accuracy in comparison to earlier research.\u003c/p\u003e"},{"header":"VI. RESULT COMPARISON WITH NORMAL VGG19","content":"\u003cp\u003eA collection of 15,000 histological images\u0026mdash;10,000 of which were of malignant lung cancer and 5,000 of which were of benign tissue\u0026mdash;were used to train and evaluate the classifier. The experimental results show that the VGG-19 architecture can achieve an accuracy rate of 95% [\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e]. Whereas our model uses over 4000 images of which approx. 2000 were benign and approx. 2000 were of malignant. By experimental findings our model can get 97.58% accuracy.\u003c/p\u003e"},{"header":"VII. CONCLUSION","content":"\u003cp\u003eDeep learning's ability to independently build features is by far its greatest advantage over other machine learning methods. Because of these characteristics, it can analyse data to identify similar properties and incorporate them for faster learning. CNNs and other deep learning techniques have transformed medical image processing in recent years. CNNs are especially well-suited for image-based medical diagnostics because of their capacity to decipher intricate visual patterns and extract valuable information from pictures. The CNN algorithm will categorise the input lung image as normal or abnormal after a successful training and testing phase. Lung cancer from histopathological scans was predicted using a Deep Convolutional Neural Network (CNN) built on the VGG-19 model. Rather than being a separate model from CNNs, VGG-19 is a specific architecture within the CNN framework. When evaluating CNN designs, performance is a crucial consideration.\u003c/p\u003e\u003cp\u003eThe experiment's findings show that the CNN with VGG-19 performs better than other CNNs. Future studies should concentrate on adding more datasets of lung cancer images and developing a more accurate model for predicting lung cancer from these images.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003eHere is the data set requested. https://www.kaggle.com/datasets/chinmay2168/lung-cancer-images\u003c/p\u003e\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eVinayak: Conceptualization, Data curation, Methodology, Software implementation, Formal analysis, Visualization, Writing \u0026ndash; original draft.Manish Madhava Tripathi: Supervision, Validation, Writing \u0026ndash; review \u0026amp; editing, Project administration, Resources.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eK. Zou et al., \u0026ldquo;National Cancer Center / National Clinical Research Center for Cancer / Cancer Yale School of Public Health, New Haven, USA,\u0026rdquo; pp. 0\u0026ndash;37, 2022, doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.jncc.2022.09.004\u003c/span\u003e\u003cspan address=\"10.1016/j.jncc.2022.09.004\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eP. R. Radhika, R. A. S. Nair, and G. Veena, \u0026ldquo;A Comparative Study of Lung Cancer Detection using Machine Learning Algorithms,\u0026rdquo; Proceedings of 2019 3rd IEEE International Conference on Electrical, Computer and Communication Technologies, ICECCT 2019, pp. 1\u0026ndash;4, 2019, doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1109/ICECCT.2019.8869001\u003c/span\u003e\u003cspan address=\"10.1109/ICECCT.2019.8869001\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHadrien T. Gayap and Moulay A. Akhloufi, \u0026ldquo;Deep Machine Learning for Medical Diagnosis, Application to Lung Cancer Detection: A Review\u0026rdquo;, BioMedInformatics Jan 2024, \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/biomedinformatics4010015\u003c/span\u003e\u003cspan address=\"10.3390/biomedinformatics4010015\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eS. Chen, Y. Han, J. Lin, X. Zhao, and P. Kong, \u0026ldquo;Pulmonary nodule detection on chest radiographs using balanced convolutional neural network and classic candidate detection,\u0026rdquo; Artificial Intelligence in Medicine, vol. 107, no. May, p. 101881, 2020, doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.artmed.2020.101881\u003c/span\u003e\u003cspan address=\"10.1016/j.artmed.2020.101881\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eY. Gu et al., \u0026ldquo;A survey of computer-aided diagnosis of lung nodules from CT scans using deep learning,\u0026rdquo; Computers in Biology and Medicine, vol. 137, no. March, 2021, doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.compbiomed.2021.104806\u003c/span\u003e\u003cspan address=\"10.1016/j.compbiomed.2021.104806\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eS. Pang and X. U. N. Wang, \u0026ldquo;A Deep Model for Lung Cancer Type Identification by Densely Connected Convolutional Networks and Adaptive Boosting,\u0026rdquo; IEEE Access, vol. 8, pp. 4799\u0026ndash;4805, 2020, doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1109/ACCESS.2019.2962862\u003c/span\u003e\u003cspan address=\"10.1109/ACCESS.2019.2962862\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eJ. Feng, A. W. Godana, S. Liu, and D. J. Gelmecha, \u0026ldquo;DFD-Net : lung cancer detection from denoised CT scan image using deep learning,\u0026rdquo; vol. 15, no. 2, 2021.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eO. Akter, M. A. Moni, M. M. Islam, J. M. W. Quinn, and A. H. M. Kamal, \u0026ldquo;Lung cancer detection using enhanced segmentation accuracy,\u0026rdquo; no. October, 2020.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eS. Pang et al., \u0026ldquo;VGG16-T : A Novel Deep Convolutional Neural Network with Boosting to Identify Pathological Type of Lung Cancer in Early Stage by CT Images,\u0026rdquo; vol. 13, no. 1, pp. 771\u0026ndash;780, 2020.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eTripathi, M. M., Haroon, M., Khan, Z., \u0026amp; Husain, M. S. (2022). Security in digital healthcare system. Pervasive Healthcare: A Compendium of Critical Factors for Success, 217\u0026ndash;231.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eManish Madhava Tripathi, Saurabh Pandey, \u0026ldquo;Diagnosis of Diabetes using Artificial Intelligence Techniques by using Bio Medical Signal Data\u0026rdquo;, International Journal of Research and Development in Applied Science and Engineering (IJRDASE) ISSN-2454-6844, Volume 13, Issue 2, May 2017\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLiangyu Lia, Jing Yang et. al.(2024), \u0026ldquo;Enhancing lung cancer detection through hybrid features and machine learning hyperparameters optimization techniques\u0026rdquo;,\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.heliyon.2024.e26192\u003c/span\u003e\u003cspan address=\"10.1016/j.heliyon.2024.e26192\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eJ. Redmon and A. Farhadi, \u0026ldquo;YOLO9000: Better, faster, stronger,\u0026rdquo; Proceedings \u0026ndash;\u0026thinsp;30th IEEE Conference on Computer Vision and Pattern Recognition, CVPR 2017, vol. 2017-Janua, pp. 6517\u0026ndash;6525, 2017, doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1109/CVPR.2017.690\u003c/span\u003e\u003cspan address=\"10.1109/CVPR.2017.690\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eM. Toğa\u0026ccedil;ar, \u0026ldquo;Disease type detection in lung and colon cancer images using the complement approach of inefficient sets,\u0026rdquo; Computers in Biology and Medicine, vol. 137, p. 104827, 2021, doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.compbiomed.2021.104827\u003c/span\u003e\u003cspan address=\"10.1016/j.compbiomed.2021.104827\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eR. Grossman, O. Haim, S. Abramov, and B. Shofty, \u0026ldquo;Differentiating Small-Cell Lung Cancer From Non-Small-Cell Lung Cancer Brain Metastases Based on MRI Using Efficientnet and Transfer Learning Approach,\u0026rdquo; vol. 20, pp. 1\u0026ndash;7, 2021, doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1177/15330338211004919\u003c/span\u003e\u003cspan address=\"10.1177/15330338211004919\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eH. Yang et al., \u0026ldquo;Deep learning-based six-type classifier for lung cancer and mimics from histopathological whole slide images : a retrospective study,\u0026rdquo; pp. 1\u0026ndash;14, 2021.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eK. Muhammad, H. Ullah, A. Khader, and J. Saudagar, \u0026ldquo;WEENet : An Intelligent System for Diagnosing COVID-19 and Lung Cancer in IoMT Environments,\u0026rdquo; vol. 11, no. February, pp. 1\u0026ndash;13, 2022, doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3389/fonc.2021.811355\u003c/span\u003e\u003cspan address=\"10.3389/fonc.2021.811355\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eA. Lang et al., \u0026ldquo;Informatics in Medicine Unlocked Model architecture and tile size selection for convolutional neural network training for non-small cell lung cancer detection on whole slide images,\u0026rdquo; Informatics in Medicine Unlocked, vol. 28, p. 100850, 2022, doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.imu.2022.100850\u003c/span\u003e\u003cspan address=\"10.1016/j.imu.2022.100850\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eL. Techniques, \u0026ldquo;A CAD System for Lung Cancer Detection Using Hybrid Deep,\u0026rdquo; 2023.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eA. Elnakib, H. M. Amer, and F. E. Z. Abou-chadi, \u0026ldquo;Early Lung Cancer Detection Using Deep Learning Optimization,\u0026rdquo; pp. 82\u0026ndash;94.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eM. Humayun, R. Sujatha, S. N. Almuayqil, and N. Z. Jhanjhi, \u0026ldquo;A Transfer Learning Approach with a Convolutional Neural Network for the Classification of Lung Carcinoma,\u0026rdquo; pp. 1\u0026ndash;15, 2022.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eG. M. M. Alshmrani, Q. Ni, R. Jiang, H. Pervaiz, and N. M. Elshennawy, \u0026ldquo;A deep learning architecture for multi-class lung diseases classification using chest X-ray (CXR) images,\u0026rdquo; Alexandria Engineering Journal, vol. 64, pp. 923\u0026ndash;935, 2023, doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.aej.2022.10.053\u003c/span\u003e\u003cspan address=\"10.1016/j.aej.2022.10.053\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eJiang, Y., Ebrahimpour, L., Despr\u0026eacute;s, P. et al. (2025). A benchmark of deep learning approaches to predict lung cancer risk using national lung screening trial cohort. Sci Rep 15, 1736. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/s41598-024-84193-7\u003c/span\u003e\u003cspan address=\"10.1038/s41598-024-84193-7\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLi, L., Yang, J., Por, L. Y., Khan, M. S., Hamdaoui, R., Hussain, L., Iqbal, Z., Rotaru, I. M., Dobrotă, D., Aldrdery, M., \u0026amp; Omar, A. (2024). Enhancing lung cancer detection through hybrid features and machine learning hyperparameters optimization techniques. Heliyon, 10(4). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.heliyon.2024.e26192\u003c/span\u003e\u003cspan address=\"10.1016/j.heliyon.2024.e26192\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLinh, V. T. N., Kim, H., Lee, M. Y., Mun, J., Kim, Y., Jeong, B. H., Park, S. G., Kim, D. H., Rho, J., \u0026amp; Jung, H. S. (2024). 3D plasmonic hexaplex paper sensor for label-free human saliva sensing and machine learning-assisted early-stage lung cancer screening. Biosensors and Bioelectronics, 244(August 2023), 115779. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.bios.2023.115779\u003c/span\u003e\u003cspan address=\"10.1016/j.bios.2023.115779\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eGayap, H. T., \u0026amp; Akhloufi, M. A. (2024). Deep Machine Learning for Medical Diagnosis, Application to Lung Cancer Detection: A Review. BioMedInformatics, 4(1), 236\u0026ndash;284. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/biomedinformatics4010015\u003c/span\u003e\u003cspan address=\"10.3390/biomedinformatics4010015\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eJaved, R., Abbas, T., Khan, A. H., Daud, A., Bukhari, A., \u0026amp; Alharbey, R. (2024). Deep learning for lungs cancer detection: a review. In Artificial Intelligence Review (Vol. 57, Issue 8). Springer Netherlands.\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s10462-024-10807-1\u003c/span\u003e\u003cspan address=\"10.1007/s10462-024-10807-1\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eAshraf, S. F., Yin, K., Meng, C. X., Wang, Q., Wang, Q., Pu, J., \u0026amp; Dhupar, R. (2022). Predicting benign, preinvasive, and invasive lung nodules on computed tomography scans using machine learning. Journal of Thoracic and Cardiovascular Surgery, 163(4), 1496\u0026ndash;1505.e10. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.jtcvs.2021.02.010\u003c/span\u003e\u003cspan address=\"10.1016/j.jtcvs.2021.02.010\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eGopinath, A., Gowthaman, P., Venkatachalam, M., \u0026amp; Saroja, M. (2023). Computer aided model for lung cancer classification using cat optimized convolutional neural networks. Measurement: Sensors, 30(January), 100932. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.measen.2023.100932\u003c/span\u003e\u003cspan address=\"10.1016/j.measen.2023.100932\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSaranya, N., Kanthimathi, N., Boomika, S., Bavatharani, S., Karthick raja, R. (2022). Classification and Prediction of Lung Cancer with Histopathological Images Using VGG-19 Architecture. In: Kalinathan, L., R., P., Kanmani, M., S., M. (eds) Computational Intelligence in Data Science. ICCIDS 2022. IFIP Advances in Information and Communication Technology, vol 654. Springer, Cham. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/978-3-031-16364-7_12\u003c/span\u003e\u003cspan address=\"10.1007/978-3-031-16364-7_12\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":true,"hasManuscriptDocX":false,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"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":"CNN architectures, CT images, Lung cancer detection, Machine learning, MMT-VIN basic model","lastPublishedDoi":"10.21203/rs.3.rs-7488263/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7488263/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eOne of the main causes of cancer-related deaths, lung cancer, requires early detection in order to be effectively treated. Our proposal is an automated method that evaluates CT data using machine learning to assist radiologists in accurately identifying troublesome lung nodules. The method focuses on nodule features via pre-processing CT images. Annotated CT scans are mostly used to train machine learning models, like CNNs, to recognize patterns of benign and cancerous nodules. To find the best effective model for detection, several CNN architectures are tested, such as ResNet, DarkNet, and EfficientNet. We present the MMT-VIN Basic model, which is based on the VGG-19 methodology. With an accuracy of almost 97.58%, the suggested lung cancer detection method outperforms the techniques it was compared to.\u003c/p\u003e","manuscriptTitle":"A Data-Driven Approach to Detecting Lung Cancer with Smart VGG19 Algorithm","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-09-12 15:12:57","doi":"10.21203/rs.3.rs-7488263/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":"358ba592-62b3-46b2-a5ba-a5bcc733cc42","owner":[],"postedDate":"September 12th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":54463156,"name":"Biological sciences/Cancer"},{"id":54463157,"name":"Biological sciences/Computational biology and bioinformatics"},{"id":54463158,"name":"Physical sciences/Engineering"},{"id":54463159,"name":"Physical sciences/Mathematics and computing"},{"id":54463160,"name":"Health sciences/Medical research"}],"tags":[],"updatedAt":"2025-09-29T08:57:51+00:00","versionOfRecord":[],"versionCreatedAt":"2025-09-12 15:12:57","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-7488263","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7488263","identity":"rs-7488263","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}