Transfer Learning with CNNs for Efficient Prostate Cancer and BPH Detection in Transrectal Ultrasound Images

Transfer Learning with CNNs for Efficient Prostate Cancer and BPH Detection in Transrectal Ultrasound Images | 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 Transfer Learning with CNNs for Efficient Prostate Cancer and BPH Detection in Transrectal Ultrasound Images Te-Li Huang, Nan-Han Lu, Yung-Hui Huang, Wen-Hung Twan, Li-Ren Yeh, and 2 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-2853191/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 09 Dec, 2023 Read the published version in Scientific Reports → Version 1 posted 11 You are reading this latest preprint version Abstract Purpose Early detection of prostate cancer (PCa) and benign prostatic hyperplasia (BPH) is crucial for maintaining the health and well-being of aging male populations. This study aims to evaluate the performance of transfer learning with convolutional neural networks (CNNs) for efficient classification of PCa and BPH in transrectal ultrasound (TRUS) images. Methods A retrospective experimental design was employed in this study, with 1,380 TRUS images for PCa and 1,530 for BPH. Seven state-of-the-art deep learning (DL) methods were employed as classifiers with transfer learning applied to popular CNN architectures. Performance indices, including sensitivity, specificity, accuracy, positive predictive value (PPV), negative predictive value (NPV), Kappa value, and Hindex (Youden's index), were used to assess the feasibility and efficacy of the CNN methods. Results The CNN methods with transfer learning demonstrated a high classification performance for TRUS images, with all accuracy, specificity, sensitivity, PPV, NPV, Kappa, and Hindex values surpassing 0.9400. The optimal accuracy, sensitivity, and specificity reached 0.9987, 0.9980, and 0.9980, respectively, as evaluated using two-fold cross-validation. Conclusion The investigated CNN methods with transfer learning showcased their efficiency and ability for the classification of PCa and BPH in TRUS images. Notably, the EfficientNetV2 with transfer learning displayed a high degree of effectiveness in distinguishing between PCa and BPH, making it a promising tool for future diagnostic applications. Prostate Cancer TRUS BPH CNN Transfer Learning Figures Figure 1 Figure 2 Introduction Deep learning methods have gained significant traction in various fields, including medicine, natural sciences, computer sciences, technical sciences, and life sciences [ 1 – 4 ]. Over the past decade, deep learning approaches have been successfully applied in a wide array of fields, such as computed tomography (CT) [ 5 – 7 ], magnetic resonance imaging (MRI) [ 9 – 10 ], digital radiography (DR) [ 10 – 12 ], positron emission tomography (PET) [ 13 – 17 ], and ultrasound tomography [ 18 – 20 ]. Given their widespread success, deep learning methods have been considered for classifying tasks within ultrasound images. One area where deep learning techniques have demonstrated promise is in the diagnosis and management of conditions that predominantly affect the aging male population. As life expectancy continues to increase globally, there is a growing need for accurate diagnostic tools and effective treatments for age-related health issues. Prostate cancer (PCa) and benign prostatic hyperplasia (BPH) are two such conditions that are highly prevalent among aging males [ 21 ]. The early detection of PCa or BPH is crucial for maintaining health and ensuring accurate diagnoses, as timely intervention can significantly improve patient outcomes. Magnetic resonance imaging (MRI), digital rectal examination (DRE), and transrectal ultrasound (TRUS) are common clinical tools for studying PCa and BPH [ 23 – 25 ]. Among these, TRUS offers several advantages, including no radiation exposure, ease of operation, and real-time scanning. However, false-positive and negative rates hinder diagnostic accuracy due to spike noise, depth attenuation effects, and scattering phenomena between media [ 26 – 32 ]. In light of these challenges, there is considerable interest in exploring the potential of deep learning techniques for improving the diagnostic process. Deep learning approaches have recently been applied to classify BPH and PCa using MRI. While TRUS remains a popular imaging tool for prostate clinical diagnostic studies, the classification of PCa and BPH using deep learning methods warrants further investigation [ 33 , 34 ]. Some studies have applied deep learning methods to diagnose prostate cancer using B-mode ultrasonography and sonoelastography [ 33 , 34 ]. B-mode ultrasound is potentially suitable for PCa imaging due to its real-time capabilities. As a result, artificial intelligence with deep learning schema has been employed to improve diagnostic accuracy in PCa via B-mode ultrasound (TRUS) [ 33 , 34 ]. Furthermore, machine learning approaches with feature-based techniques continue to progress, enhancing classification performance for PCa [ 33 , 34 ]. The application of deep learning techniques in the medical field is particularly promising due to the large volumes of data generated by modern imaging techniques. By harnessing the power of big data, researchers can develop more sophisticated algorithms capable of identifying subtle patterns and correlations that might otherwise be missed by conventional diagnostic methods. As such, the use of deep learning methods has the potential to revolutionize the way we approach the diagnosis and management of age-related health issues in the aging male population In this context, there is a pressing need for studies that systematically evaluate the performance of deep learning methods in classifying PCa and BPH using ultrasound images. Such investigations can provide valuable insights into the strengths and limitations of various techniques and help guide the development of more effective diagnostic tools. Additionally, by comparing the performance of deep learning methods with that of traditional machine learning techniques, researchers can gain a better understanding of the unique contributions that each approach brings to the table. The primary objective of this study is to investigate and compare the classification performance of PCa and BPH using B-mode ultrasound, based on popular deep learning methods with transferred learning. By exploring the potential of these advanced techniques, we aim to contribute to the ongoing efforts to improve diagnostic accuracy and early detection for both PCa and BPH, ultimately leading to better health outcomes for the aging male population. Results The testing results were evaluated based on confused matrix via two-fold cross validation. The Table 1 was shown the confused matrix provided by seven transferred learning models. All of the accuracy is over 0.95. Especially, the accuracy provided by HarDNN, MobileNetV3, ResANeT, DPResNet, and EfficientNetV2 are beyond 0.99. These results demonstrate the effectiveness of the CNN models in classifying BPH and PCa using transrectal ultrasound images. Among the seven models, EfficientNetV2 showed the highest accuracy and balanced true positive rates for both BPH and PCa. Table 1 The accuracy of classification by using seven CNN models are shown by two-fold cross validation. CNN Label BPH PCa BPH PCa Accuracy HarDNN BPH 1.000 0.000 459 0 0.9954 PCa 0.010 0.990 4 410 InceptionV3 BPH 0.950 0.050 436 23 0.9702 PCa 0.007 0.993 3 411 MobileNetV3 BPH 0.992 0.008 455 4 0.9931 PCa 0.005 0.995 2 412 CSENeT BPH 0.978 0.022 449 10 0.9771 PCa 0.024 0.976 10 404 ResANeT BPH 1.000 0.000 459 0 0.9954 PCa 0.010 0.990 4 410 DPResNet BPH 1.000 0.000 459 0 0.9920 PCa 0.017 0.983 7 407 EfficientNetV2 BPH 0.998 0.002 458 1 0.9977 PCa 0.002 0.998 1 413 Table 2 presents the performance index of the seven CNN models, including accuracy, sensitivity, specificity, PPV (positive predicted value), NPV (negative predicted value), Kappa, and Hindex, sorted in descending order by Hindex. The top-performing CNN based on the maximum Hindex value was EfficientNetV2, which provided an accuracy of 0.9977, sensitivity of 0.9980, specificity of 0.9980, PPV of 0.9976, NPV of 0.9978, and Kappa of 0.9954. The investigated CNN models demonstrated their usefulness and feasibility in performing classification between PCa and BPH sonography. Furthermore, all the investigated CNNs with transfer learning generated outstanding performance, with Hindex and Kappa values of no less than 0.940 in this study. EfficientNetV2 combined neural architecture search and scaling with added optimization. Additionally, the EfficientNetV2 method could provide adaptive regularization and dynamically adjust regularized parameters of training models according to the image size of TRUS B-mode tomography. As a result, the classification performance of EfficientNetV2 was superior to the other investigated methods in this study. Table 2 The performance index provided by seven CNN models were shown the accuracy, sensitivity, specificity, PPV (positive predicted value), NPV (negative predicted value), Kappa, and Hindex with sorting descending by Hindex. CNN Accuracy Sensitivity Specificity PPV NPV Kappa Hindex EfficientNetV2 0.9977 0.9980 0.9980 0.9976 0.9978 0.9954 0.9960 ResANeT 0.9954 0.9903 1.0000 1.0000 0.9914 0.9908 0.9903 HarDNN 0.9954 0.9900 1.0000 1.0000 0.9914 0.9908 0.9900 MobileNetV3 0.9931 0.9950 0.9920 0.9904 0.9956 0.9862 0.9870 DPResNet 0.9920 0.9831 1.0000 1.0000 0.9850 0.9839 0.9831 CSENeT 0.9771 0.9758 0.9782 0.9758 0.9782 0.9541 0.9545 InceptionV3 0.9702 0.9928 0.9499 0.9470 0.9932 0.9404 0.9431 Table 2 displays the performance index of the investigated CNNs under two-fold cross-validation with Hindex values greater than 0.940. The maximum accuracy (Acc), sensitivity (Sen), specificity (Spe), PPV, NPV, Kappa, and Hindex were generated by EfficientNetV2 with transfer learning. In contrast, the investigated CNN with the lowest Hindex was InceptionV3, which generated Acc, Sen, Spe, PPV, NPV, and Kappa values of 0.9702, 0.9928, 0.9499, 0.9470, 0.9932, and 0.9404, respectively. Therefore, feasible models should consider both low false positive and negative rates. It is challenging to minimize both false negative and positive rates simultaneously. Thus, the maximized Hindex was considered a feasible index to choose an optimal model in this study. These state-of-the-art and popular convolutional neural networks (CNN) with transfer learning were regarded as suitable for the classification task between PCa and BPH in this study (Table 2 ). The optimal investigated CNN was EfficientNetV2 in this study. Discussion Comparisions between presented results and published articles Deep learning techniques have played an essential role in diagnosing PCa over the past decade [ 28 ]. In recent articles [ 28 – 30 ], magnetic resonance imaging (MRI) has been commonly used and studied for the segmentation, classification, and detection of PCa in computer-aided diagnosis with artificial intelligence methods. However, only a few deep learning tools have been utilized for classifying PCa with TRUS images. Therefore, the primary objective of this study was to classify PCa with TRUS images and investigate the classification performance of PCa using popular CNNs with transfer learning. The analytical results showed feasible and reasonable classification between PCa and BPH (Table 2 ). The EfficientNetV2, ResANeT, HarDNN, MobileNetV3, DPResNet, CSENeT, and InceptionV3 with transfer learning are useful deep CNN methods for classifying TRUS images. Furthermore, all the accuracy, specificity, sensitivity, PPV (also called Precision), and Hindex values were higher than 0.940, as demonstrated by the investigated CNN methods in this study. The presented methods showcased their ability and efficiency. Published results using CNN or SVM are listed in Table 3 [ 31 , 42 – 45 ]. Feature-based classification between PCa and BPH with ultrasound images has commonly employed SVM classifiers [ 31 , 43 , 44 ]. However, generating useful features from images remains a challenge. In contrast, CNNs have been applied to B-mode ultrasound to create an efficient end-to-end approach [ 28 ]. Table 3 the presented method was compared with the public methods. Authors Year Classification† Method Modality Performance (%)* Zhang Q, et. al. [ 31 ] 2020 47 vs 56 SVM B-Mode US 87.9/87.0/88.8 Feng Y, et. al. [ 42 ] 2018 7511 vs 25738 cnn CEUS 90.2/82.8/91.5 Zhiyong L, et. al. [ 43 ] 2021 66 vs 103 r-cnn + xception TRUS 80.4/92.4/72.8 Huang X, et. al. [ 44 ] 2020 130 vs 126 SVM B-Mode US 70.9/70.0/71.7 Imani F, et. al. [ 45 ] 2015 625 vs 576 SVM MRI + RF-Ultrasound 80.0/88.0/80.0 This work 2022 1380 vs 1530 EfficientNetV2 TRUS 99.7/99.8/99.8 † means the sample size between malignant (positive) and benign (negative or BPH). * means accuracy/sensitivity/specificity. CEUS means the contrast-enhanced ultrasound. B-Mode US means brighten mode ultrasound. RF-Ultrasound means radio frequency ultrasound. Feature-based classification of PCa using ultrasound or contrast-enhanced ultrasound has been conducted by Zhang Q [ 31 ], Huang X [ 44 ], and Imani F [ 45 ]. The highest accuracy, sensitivity, and specificity among these results are 87.9%, 88.0%, and 88.8%. Moreover, deep learning methods have been applied to classify PCa using prostate ultrasound by Feng Y [ 42 ] and Zhiyong L [ 43 ]. The highest accuracy, sensitivity, and specificity among these results are 90.2%, 92.4%, and 91.5%. CNN improves the classification performance between PCa and benign prostate ultrasound compared with feature learning methods. The accuracy, sensitivity, and specificity achieved by EfficientNetV2 in this study are 99.4%, 99.2%, and 99.5%, respectively (Table 3 ). Transfer learning via CNNs yields satisfactory and acceptable classification results. Indeed, the methods presented in this study utilize state-of-the-art CNN techniques to create a useful model for classifying TRUS images with feasible performance. Using the state-of-art pre-trained CNNs In this study, the performance of seven state-of-the-art pre-trained CNNs was investigated for classifying TRUS images. The developed architectures of these latest CNN models were based on directed acyclic graph (DAG) networks [ 35 – 41 ]. A DAG network has a more complex architecture, designed from multiple layers. DAGs possess a rich assortment of algorithms needed for non-linear steps in the complicated geometry of multiple CNNs. A DAG network has useful properties, including reachability, transitive closure, and transitive reduction. As a result, any DAG can quickly optimize and handle multiple layers as input, as well as output from multiple layers. Meanwhile, DAG algorithms are merited for searching the shortest path for designing nodes (or architectural layers). This is one of the reasons why the investigated CNNs in this study could provide high performance classification for TRUS images. However, several parameters used in CNNs can affect the classification performance, including input image size, batch size, number of epochs, learning rates, loss function, and optimizer. Therefore, setting the parameters of a CNN model is a crucial and essential step in building a useful CNN model. Efficient Detection between BPH and PCa The success of these CNN models can be attributed to their ability to adapt to the unique characteristics of TRUS images and the utilization of transfer learning. This approach allowed the models to leverage pre-existing knowledge from large-scale datasets, enabling them to perform better in the classification task. The promising results suggest that the investigated CNNs, particularly the EfficientNetV2, hold great potential as reliable tools for accurately classifying TRUS images and distinguishing between BPH and PCa cases. Furthermore, these findings contribute to the growing body of research on the application of deep learning techniques in medical image analysis, particularly for prostate cancer diagnosis. The use of efficient and accurate deep learning models can significantly improve the clinical decision-making process and ultimately lead to better patient outcomes. By employing state-of-the-art CNNs for the classification of TRUS images, this study has demonstrated the potential of these methods in enhancing the detection of BPH and PCa, which is essential for effective treatment planning and management. Summary In this study, seven state-of-the-art CNN models were investigated for the classification of PCa and BPH using TRUS images. Early detection of prostate cancer (PCa) and benign prostatic hyperplasia (BPH) is crucial for maintaining the health and well-being of aging male populations. This paper highlights the connection between medicine and visual sciences, emphasizing the importance of utilizing advanced techniques to improve diagnostic accuracy. The modified TRUS images, with header information removed, prevented interference during the feature extraction process by the CNN models. This removal of header information from TRUS images is considered a crucial step for the successful implementation of CNN methods. With two-fold cross-validation, the accuracy, sensitivity, and specificity achieved were 0.9977, 0.9980, and 0.9980, respectively The EfficientNetV2, one of the investigated CNN methods, demonstrated its usefulness and efficiency in classifying TRUS images for differentiating PCa and BPH. Moreover, the combination of deep learning and machine learning methods has the potential to yield high classification accuracy. The presented methods can be applied to assist in the detection of TRUS images, complementing clinical diagnosis made by physicians, and ultimately contributing to better treatment planning and patient outcomes. Therefore, this study showcases the potential of state-of-the-art CNN models, particularly EfficientNetV2, in improving the classification of TRUS images for PCa and BPH. These findings have implications for enhancing the diagnostic process, leading to more timely and accurate detection, which is essential in ensuring the health and well-being of aging male populations. Limitations amd Future Works Despite the promising results, there are some limitations to this study that should be addressed and could open avenues for future work. The study used a limited dataset of TRUS images for the classification of PCa and BPH. Expanding the dataset with a larger and more diverse collection of images could further validate the effectiveness of the proposed methods and improve the generalizability of the models. While transfer learning was employed in this study, it may be beneficial to explore other pre-training and fine-tuning strategies to enhance the performance of the CNN models. Additionally, the development of task-specific architectures tailored to the classification of PCa and BPH using TRUS images could lead to even better results. Future work could explore the integration of multi-modal data, such as combining TRUS images with MRI or other diagnostic tools, to enhance the detection and classification performance of the models. Developing methods to better understand and visualize the features learned by the CNN models can provide insights into the decision-making process, potentially leading to increased trust and adoption of the models by medical professionals. Future work could investigate the development of real-time classification systems that can be integrated with existing medical imaging devices, providing immediate feedback to clinicians during patient examinations. Investigating the performance of the CNN models in longitudinal studies, where patient data is collected over time, could provide insights into the progression of PCa and BPH, as well as the effectiveness of different treatments. By addressing these limitations and exploring future work, the application of CNN models in the classification of PCa and BPH using TRUS images could be further optimized, contributing to more accurate and timely detection and improving patient outcomes. Methods and Materials Ethics approval This study was conducted after approval by the Institutional Review Board of Kaohsiung Veterans General Hospital (VGHKS IRB; No. VGHKS13-CT6-04). Due to the retrospective nature of the study, informed consent was waived by the VGHKS IRB. This study was conducted with the methods in accordance with relevant guidelines and regulations. The enrolled samples and research flowchart The cases in PCa and BPH were 1,380 and 1,530. The age (years), prostate-specific antigen (PSA (ng/mL)), and TRUS images were collected in this work. The descriptive statistics were shown in Table 4 . The mean ± SD (Standard Deviation, SD) of age between BPH and PCa groups were 64.3 ± 9.4 (years) and 68.1 ± 9.8 (years). The mean ± SD of PSA between BPH and PCa were 1.7 ± 0.5 (ng/mL) and 28.7 ± 45.7 (ng/mL). All of the collected TRUS images were confirmed by pathological biopsy to identify as PCa or BPH. Table 4 the age and PSA between groups. Group Age (years) PSA (ng/mL) Mean STD Mean STD BPH(n = 1,530) 64.3 9.4 1.7 0.5 PCa (n = 1,380) 68.1 9.8 28.7 45.7 The preprocessing of the TRUS images was necessary before training classification models. In order to remove any identifying information (such as patient name, patient ID, hospital name, and other information) from the input images, the modified images were created by setting intensity zeros outside the prostate area (Fig. 1 ). The TRUS images were in the size of 640 × 480 with gray-level PNG format. The size and format of the modified images remained the same as the input images. Meanwhile, the extra information might interfere with the classification accuracy using the presented methods. The modified images were created according to the boundary of the dashed line in the input image. The main purpose of creating modified images was to avoid the influences on classification due to the header information around the image (Fig. 1 right). The header information included patient ID, study date, and scanning parameters (i.e., depth ruler, imaging settings, gray level bar, and measurement information). The flowchart in this study included input images, preprocessing, pre-training deep learning methods (DL) via transfer learning, validation, and results (Fig. 2 ). In the image processing step, the input images were modified by excluding header information and saved as new images. The investigated popular and latest deep learning approaches included seven deep learning methods, as described in Section 2.2. Next, the investigated models with different parameter settings were validated based on accuracy, specificity, sensitivity, negative predictive value (NPV), positive predictive value (PPV), Kappa, and Hindex values. The Deep Learning via Transferred Learning The DL methods employed the popular convolutional neural networks (CNN), including feature-map based convolutional neural network (HarDNN) [ 35 ], InceptionV3 [ 36 ], MobileNetV3 [ 37 ], competitive squeeze and excitation neural network (CSENeT) [ 38 ], residual attention neural network (ResANeT) [ 39 ], deep pyramidal residual neural network (DPResNet) [ 40 ], and EfficientNetV2 [ 41 ] via transfer learning (Table 5 ). The investigated CNN methods were regarded as the best deep CNN architectures for classification tasks. Meanwhile, the classification layer and fully connected layer were modified and fixed for two classes (i.e., the original designed number of classes was 1,000). Additionally, a batch size of 5 was investigated for the optimal accuracy of the presented methods. The two-fold cross-validation was designed to evaluate the classification performance for the presented methods. Table 5 The merits of investigated CNNs in recently. CNN Authors Year Merits HarDNN Mahmoud et. al. [ 35 ] 2020 The HarDNN was accurately estimated a feature map in CNN. HarDNN was demonstrated to improve computational times by 10x faster than other CNNs. InceptionV3 Kin N. [ 36 ] 2019 The InceptionV3 was proposed to prompt the accuracy of recognition by performing a fine-tuned parameter in the layers of CNN with transferred learning pre-trained model. The InceptionV3 was trained and validated on a cooking dataset. The InceptionV3 was provide a potential solution to the recognition problem. MobileNetV3 Chu et. al. [ 37 ] 2019 MobileNet was used the depth-wise separable convolutions to create a small weight CNN. The global hyper-parameters of MobileNet that efficiently balance between latent parameters and classification accuracy by evaluating ImageNet classification. Meanwhile, MobileNet had demonstrated the across a wide range of applications including object detection, fine-grain classification, face attributes and large-scale geo-localization. CSENeT Hu et. al. [ 38 ] 2018 The competitive squeeze-excitation residual network (CSENet) was built for to determined and to expand the meaning of channel relationship in residual layers. The CSENet was proven as good as the popular CNN. ResANeT Wang et. al. [ 39 ] 2017 The residual attention network (ResANeT) was built according stacking attention modules with generating attention-aware features. The ResANeT not only be used to train deep residual attention network, but could be easily adjusted to hundreds of layers. DPResNet Han et.al. [ 40 ] 2016 The deep pyramidal residual neural network (DPResNet) was designed to ensure effective performance by increasing the diversity of high-level features in an image. The DPResNet had been proven to improve generalization ability to prompt the classification accuracy. EfficientNetV2 Tan et.al. [ 41 ] 2021 The EfficientNetV2 was a new CNN that trains faster and efficiency parameters than other CNN models. The combination of neural architecture searching and scaling with adding optimization. The EfficientNetV2 significantly outperforms previous models on ImageNet and CIFAR/Cars/Flowers datasets. Evaluated Perofrmance of Presented Methods The performance of the presented models was evaluated using the testing set. The testing set consisted of 50% randomly sampled data from each group, with two-fold cross-validation. The testing performance of the presented methods is typically assessed using popular indices. A confusion matrix is often employed in the literature to evaluate the suitability of different models, including their sensitivity, specificity, accuracy, positive predictive value (PPV), negative predictive value (NPV), Kappa value, and Hindex index (or Youden's index). Abbreviations The abbreviations used in this text are listed with their full names below. Abbreviation Full Name AI Artificial intelligent BPH Benign prostatic hyperplasia CNN Convolutional neuron networks CSENeT Competitive squeeze and excitation neural network CT Computed tomography DAG Directed acyclic graph DL Deep learning DPResNet Deep pyramidal residual neural network DR Digital radiography HarDNN Feature-map based convolutional neural network Hindex Youden's index MRI Magnetic resonance imaging NPV Negative predictive value patient ID Patient identification PCa Prostate cancer PET Positron emission tomography PPV Positive predictive value or Precision PSA Prostate-specific antigen ResANeT Residual attention neural network STD Standard Deviation TRUS Transrectal ultrasound Declarations Author Contributions: Conceptualization, T.L.H. and T.B.C.; methodology, N.H.L., Y.H.H, and T.B.C; software, W.H.T.; validation, N.H.L. and T.B.C.; formal analysis, T.L.H.; investigation, T.L.H.; resources, Y.H.H; data curation T.L.H. and K.Y.L; writing—original draft preparation, T.L.H., T.B.C. and N.H.L.; writing—review and editing, T.B.C. and N.H.L.; visualization, L.R.Y. and K.YL; supervision, T.B.C.; project administration T.B.C. All authors have read and agreed to the published version of the manuscript. Funding: The authors would like to thank the National Science and Technology Council in Taiwan for financially supporting this research under contract NSTC 111-2118-M-214-001 and 110-2118-M-214-001. Competing interests: The authors declare no conflict of interest. Ethics approval and consent to participate Consent for publication: This study was conducted after approval by the Institutional Review Board of Kaohsiung Veterans General Hospital (VGHKS IRB; No. VGHKS13-CT6-04). Due to the retrospective nature of the study, informed consent was waived by the VGHKS IRB. This study was conducted with the methods in accordance with relevant guidelines and regulations. Availability of data and material: The data presented in this study are available upon request from the corresponding author. The data are not publicly available due to restrictions, e.g., privacy and ethical concerns. Acknowledgments: This research was partially supported by grants from the National Science and Technology Council, R.O.C (NSTC110-2118-M-214-001, and NSTC 111-2118-M-214-001). References Morgan N, Van Gerven A, Smolders A, de Faria Vasconcelos K, Willems H, Jacobs R. Convolutional neural network for automatic maxillary sinus segmentation on cone-beam computed tomographic images. Sci Rep. 2022 May 7;12(1):7523. doi:10.1038/s41598-022-11483-3 Deng Y, Wang L, Zhao C, Tang S, Cheng X, Deng HW, Zhou W. A deep learning-based approach to automatic proximal femur segmentation in quantitative CT images. Med Biol Eng Comput. 2022 May;60(5):1417-1429. doi:10.1007/s11517-022-02529-9. Hung KF, Ai QYH, King AD, Bornstein MM, Wong LM, Leung YY. Automatic detection and segmentation of morphological changes of the maxillary sinus mucosa on cone-beam computed tomography images using a three-dimensional convolutional neural network. Clin Oral Investig. 2022 May;26(5):3987-3998. doi:10.1007/s00784-021-04365-x. Valjarević, A.; Morar, C.; Živković, J.; Niemets, L.; Kićović, D.; Golijanin, J.; Gocić, M.; Bursać, N.M.; Stričević, L.; Žiberna, I.; Bačević, N.; Milevski, I.; Durlević, U.; Lukić, T. Long Term Monitoring and Connection between Topography and Cloud Cover Distribution in Serbia. Atmosphere 2021, 12, 964. Doi: 10.3390/atmos12080964. Berhane H, Scott MB, Barker AJ, McCarthy P, Avery R, Allen B, Malaisrie C, Robinson JD, Rigsby CK, Markl M. Deep learning-based velocity antialiasing of 4D-flow MRI. Magn Reson Med. 2022 Jul;88(1):449-463. doi: 10.1002/mrm.29205. Luciw NJ, Shirzadi Z, Black SE, Goubran M, MacIntosh BJ. Automated generation of cerebral blood flow and arterial transit time maps from multiple delay arterial spin-labeled MRI. Magn Reson Med. 2022 Jul;88(1):406-417. doi:10.1002/mrm.29193. Arabahmadi M, Farahbakhsh R, Rezazadeh J. Deep Learning for Smart Healthcare-A Survey on Brain Tumor Detection from Medical Imaging. Sensors(Basel). 2022 Mar 2;22(5):1960. doi: 10.3390/s22051960. Aggarwal P, Mishra NK, Fatimah B, Singh P, Gupta A, Joshi SD. COVID-19 image classification using deep learning: Advances, challenges and opportunities. Comput Biol Med. 2022 May;144:105350. doi:10.1016/j.compbiomed.2022.105350. Aggarwal P, Mishra NK, Fatimah B, Singh P, Gupta A, Joshi SD. COVID-19 image classification using deep learning: Advances, challenges and opportunities. Comput Biol Med. 2022 May;144:105350. doi: 10.1016/j.compbiomed.2022.105350. Mulrenan C, Rhode K, Fischer BM. A Literature Review on the Use of Artificial Intelligence for the Diagnosis of COVID-19 on CT and Chest X-ray. Diagnostics (Basel). 2022 Mar 31;12(4):869. doi: 10.3390/diagnostics12040869. Minoshima S, Cross D. Application of artificial intelligence in brain molecular imaging. Ann Nucl Med. 2022 Feb;36(2):103-110. doi:10.1007/s12149-021-01697-2. Xue S, Guo R, Bohn KP, Matzke J, Viscione M, Alberts I, Meng H, Sun C, Zhang M, Zhang M, Sznitman R, El Fakhri G, Rominger A, Li B, Shi K. A cross-scanner and cross-tracer deep learning method for the recovery of standard-dose imaging quality from low-dose PET. Eur J Nucl Med Mol Imaging. 2022 May;49(6):1843-1856. doi:10.1007/s00259-021-05644-1. Dal Toso L, Chalampalakis Z, Buvat I, Comtat C, Cook G, Goh V, Schnabel JA, Marsden PK. Improved 3D tumour definition and quantification of uptake in simulated lung tumours using deep learning. Phys Med Biol. 2022 Apr 27;67(9). doi:10.1088/1361-6560/ac65d6. Mamistvalov A, Amar A, Kessler N, Eldar YC. Deep-Learning Based Adaptive Ultrasound Imaging From Sub-Nyquist Channel Data. IEEE Trans Ultrason Ferroelectr Freq Control. 2022 May;69(5):1638-1648. doi:10.1109/TUFFC.2022.3160859. Sato M, Kobayashi T, Soroida Y, Tanaka T, Nakatsuka T, Nakagawa H, Nakamura A, Kurihara M, Endo M, Hikita H, Sato M, Gotoh H, Iwai T, Tateishi R, Koike K, Yatomi Y. Development of novel deep multimodal representation learning-based model for the differentiation of liver tumors on B-mode ultrasound images. J Gastroenterol Hepatol. 2022 Apr;37(4):678-684. doi: 10.1111/jgh.15763. Song Y, Zheng J, Lei L, Ni Z, Zhao B, Hu Y. CT2US: Cross-modal transfer learning for kidney segmentation in ultrasound images with synthesized data. Ultrasonics. 2022 May;122:106706. doi: 10.1016/j.ultras.2022.106706. Huang X, Chen M, Liu P, Du Y. Texture Feature-Based Classification on Transrectal Ultrasound Image for Prostatic Cancer Detection. Comput Math Methods Med. 2020 Oct 6;2020:7359375. doi: 10.1155/2020/7359375. Aleksandar V, Tatjana D, Vladica S, Radomir I, Bojana J, GIS numerical and remote sensing analyses of forest changes in the Toplica region for the period of 1953–2013. Applied Geography. 2018;92:131-139. doi:10.1016/j.apgeog.2018.01.016. Ahmad, J., Farman, H., Jan, Z. (2019). Deep Learning Methods and Applications. In: Deep Learning: Convergence to Big Data Analytics. Springer Briefs in Computer Science. Springer, Singapore. Doi: 10.1007/978-981-13-3459-7_3. Faure Walker N, Gill B, Olsburgh J, Gillatt D, Yap T, Michala L, Taylor C, Wood H, Wood D. Age-related urologic problems in the complex urologic patient. World J Urol. 2021 Apr;39(4):1037-1044. doi:10.1007/s00345-020-03111-4. Sedghi A, Pesteie M, Javadi G, Azizi S, Yan P, Kwak JT, Xu S, Turkbey B, Choyke P, Pinto P, Wood B, Rohling R, Abolmaesumi P, Mousavi P. Deep neural maps for unsupervised visualization of high-grade cancer in prostate biopsies. Int J Comput Assist Radiol Surg. 2019 Jun;14(6):1009-1016. doi:10.1007/s11548-019-01950-0. Wasserman NF, Niendorf E, Spilseth B. Precision and accuracy of magnetic resonance imaging for lobar classification of benign prostatic hyperplasia. Abdom Radiol (NY). 2019 Jul;44(7):2535-2544. doi:10.1007/s00261-019-01970-z. Wang X, Yang W, Weinreb J, Han J, Li Q, Kong X, Yan Y, Ke Z, Luo B, Liu T, Wang L. Searching for prostate cancer by fully automated magnetic resonance imaging classification: deep learning versus non-deep learning. Sci Rep. 2017 Nov 13;7(1):15415. doi: 10.1038/s41598-017-15720-y. Song Y, Zhang YD, Yan X, Liu H, Zhou M, Hu B, Yang G. Computer-aided diagnosis of prostate cancer using a deep convolutional neural network from multiparametric MRI. J Magn Reson Imaging. 2018 Dec;48(6):1570-1577. doi:10.1002/jmri.26047. Aldoj N, Lukas S, Dewey M, Penzkofer T. Semi-automatic classification of prostate cancer on multi-parametric MR imaging using a multi-channel 3D convolutional neural network. Eur Radiol. 2020 Feb;30(2):1243-1253. doi:10.1007/s00330-019-06417-z. Yuan Y, Qin W, Buyyounouski M, Ibragimov B, Hancock S, Han B, Xing L. Prostate cancer classification with multiparametric MRI transfer learning model. Med Phys. 2019 Feb;46(2):756-765. doi: 10.1002/mp.13367. Wildeboer RR, van Sloun RJG, Wijkstra H, Mischi M. Artificial intelligence in multiparametric prostate cancer imaging with focus on deep-learning methods. Comput Methods Programs Biomed. 2020 Jun;189:105316. doi:10.1016/j.cmpb.2020.105316. Tătaru OS, Vartolomei MD, Rassweiler JJ, Virgil O, Lucarelli G, Porpiglia F, Amparore D, Manfredi M, Carrieri G, Falagario U, Terracciano D, de Cobelli O, Busetto GM, Del Giudice F, Ferro M. Artificial Intelligence and Machine Learning in Prostate Cancer Patient Management-Current Trends and Future Perspectives. Diagnostics (Basel). 2021 Feb 20;11(2):354. Syer T, Mehta P, Antonelli M, Mallett S, Atkinson D, Ourselin S, Punwani S. Artificial Intelligence Compared to Radiologists for the Initial Diagnosis of Prostate Cancer on Magnetic Resonance Imaging: A Systematic Review and Recommendations for Future Studies. Cancers (Basel). 2021 Jul 1;13(13):3318. doi: 10.3390/cancers13133318. Zhang Q, Xiong J, Cai Y, Shi J, Xu S, Zhang B. Multimodal feature learning and fusion on B-mode ultrasonography and sonoelastography using point-wise gated deep networks for prostate cancer diagnosis. Biomed Tech (Berl). 2020 Jan 28;65(1):87-98. doi: 10.1515/bmt-2018-0136. Wildeboer RR, Mannaerts CK, van Sloun RJG, Budäus L, Tilki D, Wijkstra H, Salomon G, Mischi M. Automated multiparametric localization of prostate cancer based on B-mode, shear-wave elastography, and contrast-enhanced ultrasound radiomics. Eur Radiol. 2020 Feb;30(2):806-815. doi: 10.1007/s00330-019-06436-w. Shah M, Naik N, Somani BK, Hameed BMZ. Artificial intelligence (AI) in urology-Current use and future directions: An iTRUE study. Turk J Urol. 2020 Nov;46(Supp. 1):S27-S39. doi: 10.5152/tud.2020.20117. Chiu PK, Shen X, Wang G, Ho CL, Leung CH, Ng CF, Choi KS, Teoh JY. Enhancement of prostate cancer diagnosis by machine learning techniques: an algorithm development and validation study. Prostate Cancer Prostatic Dis. 2021 Jul 15. doi: 10.1038/s41391-021-00429-x. Mahmoud A, Hari S, Fletcher CW, Adve SV, Sakr C, Shanbhag N, Molchanov P, Sullivan MB, Tsai T, Keckler SW. HarDNN: Feature Map Vulnerability Evaluation in CNNs. 2020 Feb; arXiv:2002.09786. doi:10.48550/arXiv.2002.09786, Kin N. Tuned Inception V3 for Recognizing States of Cooking Ingredients. 2019 May; arXiv:1905.03715. doi: 10.48550/arXiv.1905.03715. Chu X, Zhang B, Xu R. MoGA: Searching Beyond MobileNetV3. 2019 Aug; arXiv:1908.01314. doi: 10.48550/arXiv.1908.01314. Hu Y, Wen G, Luo M, Dai D, Ma J, Yu Z. Competitive Inner-Imaging Squeeze and Excitation for Residual Network. 2018 Jul; arXiv:1807.08920. doi:10.48550/arXiv.1807.08920. Wang F, Jiang M, Qian C, Yang S, Li C, Zhang H, Wang X, Tang X. Residual Attention Network for Image Classification. 2017 Apr; arXiv:1704.06904. doi:10.48550/arXiv.1704.06904. Han D, Kim J, Kim J. Deep Pyramidal Residual Networks. 2019 Oct; arXiv:1610.02915. doi:10.48550/arXiv.1610.02915. Tan M, Le QV. EfficientNetV2: Smaller Models and Faster Training. 2021 Apr; arXiv:2104.00298. doi:10.48550/arXiv.2104.00298 Feng Y, Yang F, Zhou X, Guo Y, Tang F, Ren F, Guo J, Ji S. A Deep Learning Approach for Targeted Contrast-Enhanced Ultrasound Based Prostate Cancer Detection. IEEE/ACM Trans Comput Biol Bioinform. 2019 Nov-Dec;16(6):1794-1801. doi: 10.1109/TCBB.2018.2835444. Zhiyong L, Chuan Y, Jun H, Shaopeng L, Yumin Z. Xu L. Deep learning framework based on integration of S-Mask R-CNN and Inception-v3 for ultrasound image-aided diagnosis of prostate cancer. Future Generation Computer Systems. 2021 Jan; 114:258-367. doi:10.1016/j.future.2020.08.015. Huang X, Chen M, Liu P, Du Y. Texture Feature-Based Classification on Transrectal Ultrasound Image for Prostatic Cancer Detection. Comput Math Methods Med. 2020 Oct 6;2020:7359375. doi: 10.1155/2020/7359375. Imani F, Abolmaesumi P, Gibson E, Khojaste A, Gaed M, Moussa M, Gomez JA, Romagnoli C, Leveridge M, Chang S, Siemens DR, Fenster A, Ward AD, Mousavi P. Computer-Aided Prostate Cancer Detection Using Ultrasound RF Time Series: In Vivo Feasibility Study. IEEE Trans Med Imaging. 2015 Nov;34(11):2248-57. doi:10.1109/TMI.2015.2427739. Additional Declarations No competing interests reported. Cite Share Download PDF Status: Published Journal Publication published 09 Dec, 2023 Read the published version in Scientific Reports → Version 1 posted Editorial decision: Major revision 17 Aug, 2023 Reviews received at journal 16 Aug, 2023 Reviews received at journal 09 Aug, 2023 Reviews received at journal 28 Jul, 2023 Reviewers agreed at journal 19 Jul, 2023 Reviewers agreed at journal 19 Jul, 2023 Reviewers invited by journal 18 Jul, 2023 Editor assigned by journal 18 Jul, 2023 Editor invited by journal 10 May, 2023 Submission checks completed at journal 10 May, 2023 First submitted to journal 24 Apr, 2023 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-2853191","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":198840931,"identity":"9a02a77c-c1be-44c2-bb83-6c19dac91529","order_by":0,"name":"Te-Li Huang","email":"","orcid":"","institution":"Kaohsiung Veterans General Hospital","correspondingAuthor":false,"submittingAuthor":false,"prefix":"","firstName":"Te-Li","middleName":"","lastName":"Huang","suffix":""},{"id":198840932,"identity":"87f223b8-72a3-4cef-ace0-5de7ed2b5ead","order_by":1,"name":"Nan-Han Lu","email":"","orcid":"","institution":"E-DA Cancer Hospital, I-Shou University","correspondingAuthor":false,"submittingAuthor":false,"prefix":"","firstName":"Nan-Han","middleName":"","lastName":"Lu","suffix":""},{"id":198840933,"identity":"9eb3e3ad-81ea-4df9-a2d1-6a9f97f61208","order_by":2,"name":"Yung-Hui Huang","email":"","orcid":"","institution":"I-Shou University","correspondingAuthor":false,"submittingAuthor":false,"prefix":"","firstName":"Yung-Hui","middleName":"","lastName":"Huang","suffix":""},{"id":198840934,"identity":"d3b628d2-5902-42ec-93af-cfbe1ec76d39","order_by":3,"name":"Wen-Hung Twan","email":"","orcid":"","institution":"National Taitung University","correspondingAuthor":false,"submittingAuthor":false,"prefix":"","firstName":"Wen-Hung","middleName":"","lastName":"Twan","suffix":""},{"id":198840935,"identity":"25b0492a-11a2-47e2-a562-db37186bbffb","order_by":4,"name":"Li-Ren Yeh","email":"","orcid":"","institution":"E-DA Cancer Hospital, I-Shou University","correspondingAuthor":false,"submittingAuthor":false,"prefix":"","firstName":"Li-Ren","middleName":"","lastName":"Yeh","suffix":""},{"id":198840936,"identity":"4bb8765f-6556-48cc-84fd-78fd87c76188","order_by":5,"name":"Kuo-Ying Liu","email":"","orcid":"","institution":"E-DA Cancer Hospital, I-Shou University","correspondingAuthor":false,"submittingAuthor":false,"prefix":"","firstName":"Kuo-Ying","middleName":"","lastName":"Liu","suffix":""},{"id":198840938,"identity":"d8c0595e-d8a4-42ed-b309-65983d7011f5","order_by":6,"name":"Tai-Been Chen","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAApklEQVRIiWNgGAWjYFACHsYHCWBGArE62HiYDeBaDhCphU2CgSQt/PN7j1U8+HOYgZ89x4D5YxsRWiSO8aXdSGw7zCDZ88aA4SAxWgzYeMxuJDYcZjC4kQPUso1ILQUJQIfZk6SFIYENaIsEsVokjuUlSyS2pfNInHlWcODsPyK08DefPfjxxx9rOf725I0PKs4QoQUGeEDEARI0jIJRMApGwSjABwA6gjT9xTTe2QAAAABJRU5ErkJggg==","orcid":"","institution":"I-Shou University","correspondingAuthor":true,"submittingAuthor":false,"prefix":"","firstName":"Tai-Been","middleName":"","lastName":"Chen","suffix":""}],"badges":[],"createdAt":"2023-04-24 06:44:23","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-2853191/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-2853191/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1038/s41598-023-49159-1","type":"published","date":"2023-12-09T15:01:42+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":36958439,"identity":"499219bd-4117-4709-ac2d-a593ffe0c720","added_by":"auto","created_at":"2023-05-12 14:50:52","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":364975,"visible":true,"origin":"","legend":"\u003cp\u003eshows the (A) maligned prostate cancer and (B) BPH images.\u003c/p\u003e","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-2853191/v1/5ba2b2f9ba523e4567a5c5b3.png"},{"id":36959680,"identity":"4bdb9b1e-a87b-43b0-8ef1-3219b918cb0c","added_by":"auto","created_at":"2023-05-12 14:58:52","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":60580,"visible":true,"origin":"","legend":"\u003cp\u003eshow the flowchart in this study.\u003c/p\u003e","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-2853191/v1/c5624bd4ff7ec9eb541a432b.png"},{"id":47989297,"identity":"07a26c43-1ead-4872-a3cf-b0f048bd44bb","added_by":"auto","created_at":"2023-12-11 15:08:47","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":823804,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-2853191/v1/410baf32-05af-4875-90f7-0e779e4aa51e.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Transfer Learning with CNNs for Efficient Prostate Cancer and BPH Detection in Transrectal Ultrasound Images","fulltext":[{"header":"Introduction","content":"\u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eDeep learning methods have gained significant traction in various fields, including medicine, natural sciences, computer sciences, technical sciences, and life sciences [\u003cspan additionalcitationids=\"CR2 CR3\" citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e]. Over the past decade, deep learning approaches have been successfully applied in a wide array of fields, such as computed tomography (CT) [\u003cspan additionalcitationids=\"CR6\" citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e], magnetic resonance imaging (MRI) [\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e], digital radiography (DR) [\u003cspan additionalcitationids=\"CR11\" citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e], positron emission tomography (PET) [\u003cspan additionalcitationids=\"CR14 CR15 CR16\" citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e], and ultrasound tomography [\u003cspan additionalcitationids=\"CR19\" citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e]. Given their widespread success, deep learning methods have been considered for classifying tasks within ultrasound images.\u003c/p\u003e \u003cp\u003eOne area where deep learning techniques have demonstrated promise is in the diagnosis and management of conditions that predominantly affect the aging male population. As life expectancy continues to increase globally, there is a growing need for accurate diagnostic tools and effective treatments for age-related health issues. Prostate cancer (PCa) and benign prostatic hyperplasia (BPH) are two such conditions that are highly prevalent among aging males [\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e]. The early detection of PCa or BPH is crucial for maintaining health and ensuring accurate diagnoses, as timely intervention can significantly improve patient outcomes.\u003c/p\u003e \u003cp\u003eMagnetic resonance imaging (MRI), digital rectal examination (DRE), and transrectal ultrasound (TRUS) are common clinical tools for studying PCa and BPH [\u003cspan additionalcitationids=\"CR24\" citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e]. Among these, TRUS offers several advantages, including no radiation exposure, ease of operation, and real-time scanning. However, false-positive and negative rates hinder diagnostic accuracy due to spike noise, depth attenuation effects, and scattering phenomena between media [\u003cspan additionalcitationids=\"CR27 CR28 CR29 CR30 CR31\" citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e]. In light of these challenges, there is considerable interest in exploring the potential of deep learning techniques for improving the diagnostic process.\u003c/p\u003e \u003cp\u003eDeep learning approaches have recently been applied to classify BPH and PCa using MRI. While TRUS remains a popular imaging tool for prostate clinical diagnostic studies, the classification of PCa and BPH using deep learning methods warrants further investigation [\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e, \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e]. Some studies have applied deep learning methods to diagnose prostate cancer using B-mode ultrasonography and sonoelastography [\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e, \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e]. B-mode ultrasound is potentially suitable for PCa imaging due to its real-time capabilities. As a result, artificial intelligence with deep learning schema has been employed to improve diagnostic accuracy in PCa via B-mode ultrasound (TRUS) [\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e, \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e]. Furthermore, machine learning approaches with feature-based techniques continue to progress, enhancing classification performance for PCa [\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e, \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eThe application of deep learning techniques in the medical field is particularly promising due to the large volumes of data generated by modern imaging techniques. By harnessing the power of big data, researchers can develop more sophisticated algorithms capable of identifying subtle patterns and correlations that might otherwise be missed by conventional diagnostic methods. As such, the use of deep learning methods has the potential to revolutionize the way we approach the diagnosis and management of age-related health issues in the aging male population\u003c/p\u003e \u003cp\u003eIn this context, there is a pressing need for studies that systematically evaluate the performance of deep learning methods in classifying PCa and BPH using ultrasound images. Such investigations can provide valuable insights into the strengths and limitations of various techniques and help guide the development of more effective diagnostic tools. Additionally, by comparing the performance of deep learning methods with that of traditional machine learning techniques, researchers can gain a better understanding of the unique contributions that each approach brings to the table.\u003c/p\u003e \u003cp\u003eThe primary objective of this study is to investigate and compare the classification performance of PCa and BPH using B-mode ultrasound, based on popular deep learning methods with transferred learning. By exploring the potential of these advanced techniques, we aim to contribute to the ongoing efforts to improve diagnostic accuracy and early detection for both PCa and BPH, ultimately leading to better health outcomes for the aging male population.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e"},{"header":"Results","content":"\u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eThe testing results were evaluated based on confused matrix via two-fold cross validation. The Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e was shown the confused matrix provided by seven transferred learning models. All of the accuracy is over 0.95. Especially, the accuracy provided by HarDNN, MobileNetV3, ResANeT, DPResNet, and EfficientNetV2 are beyond 0.99. These results demonstrate the effectiveness of the CNN models in classifying BPH and PCa using transrectal ultrasound images. Among the seven models, EfficientNetV2 showed the highest accuracy and balanced true positive rates for both BPH and PCa.\u003c/p\u003e \u003c/div\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\u003eThe accuracy of classification by using seven CNN models are shown by two-fold cross validation.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"7\"\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=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCNN\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eLabel\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eBPH\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003ePCa\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eBPH\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003ePCa\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e \u003cp\u003eAccuracy\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eHarDNN\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eBPH\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e1.000\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.000\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e459\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003e0.9954\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003ePCa\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.010\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.990\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e410\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eInceptionV3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eBPH\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.950\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.050\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e436\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e23\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003e0.9702\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003ePCa\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.007\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.993\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e411\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eMobileNetV3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eBPH\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.992\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.008\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e455\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003e0.9931\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003ePCa\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.005\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.995\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e412\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eCSENeT\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eBPH\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.978\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.022\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e449\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e10\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003e0.9771\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003ePCa\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.024\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.976\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e10\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e404\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eResANeT\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eBPH\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e1.000\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.000\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e459\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003e0.9954\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003ePCa\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.010\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.990\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e410\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eDPResNet\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eBPH\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e1.000\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.000\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e459\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003e0.9920\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003ePCa\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.017\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.983\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e7\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e407\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eEfficientNetV2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eBPH\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.998\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.002\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e458\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003e0.9977\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003ePCa\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.002\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.998\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e413\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=\"BlockQuote\"\u003e \u003cp\u003eTable\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e presents the performance index of the seven CNN models, including accuracy, sensitivity, specificity, PPV (positive predicted value), NPV (negative predicted value), Kappa, and Hindex, sorted in descending order by Hindex. The top-performing CNN based on the maximum Hindex value was EfficientNetV2, which provided an accuracy of 0.9977, sensitivity of 0.9980, specificity of 0.9980, PPV of 0.9976, NPV of 0.9978, and Kappa of 0.9954. The investigated CNN models demonstrated their usefulness and feasibility in performing classification between PCa and BPH sonography. Furthermore, all the investigated CNNs with transfer learning generated outstanding performance, with Hindex and Kappa values of no less than 0.940 in this study. EfficientNetV2 combined neural architecture search and scaling with added optimization. Additionally, the EfficientNetV2 method could provide adaptive regularization and dynamically adjust regularized parameters of training models according to the image size of TRUS B-mode tomography. As a result, the classification performance of EfficientNetV2 was superior to the other investigated methods in this study.\u003c/p\u003e \u003c/div\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\u003eThe performance index provided by seven CNN models were shown the accuracy, sensitivity, specificity, PPV (positive predicted value), NPV (negative predicted value), Kappa, and Hindex with sorting descending by Hindex.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"8\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c8\" colnum=\"8\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCNN\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAccuracy\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eSensitivity\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eSpecificity\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003ePPV\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eNPV\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e \u003cp\u003eKappa\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c8\"\u003e \u003cp\u003eHindex\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eEfficientNetV2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.9977\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.9980\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.9980\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.9976\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.9978\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.9954\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.9960\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eResANeT\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.9954\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.9903\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e1.0000\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e1.0000\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.9914\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.9908\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.9903\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHarDNN\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.9954\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.9900\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e1.0000\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e1.0000\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.9914\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.9908\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.9900\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMobileNetV3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.9931\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.9950\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.9920\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.9904\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.9956\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.9862\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.9870\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDPResNet\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.9920\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.9831\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e1.0000\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e1.0000\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.9850\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.9839\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.9831\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCSENeT\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.9771\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.9758\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.9782\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.9758\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.9782\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.9541\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.9545\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eInceptionV3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.9702\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.9928\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.9499\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.9470\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.9932\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.9404\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.9431\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=\"BlockQuote\"\u003e \u003cp\u003eTable\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e displays the performance index of the investigated CNNs under two-fold cross-validation with Hindex values greater than 0.940. The maximum accuracy (Acc), sensitivity (Sen), specificity (Spe), PPV, NPV, Kappa, and Hindex were generated by EfficientNetV2 with transfer learning. In contrast, the investigated CNN with the lowest Hindex was InceptionV3, which generated Acc, Sen, Spe, PPV, NPV, and Kappa values of 0.9702, 0.9928, 0.9499, 0.9470, 0.9932, and 0.9404, respectively. Therefore, feasible models should consider both low false positive and negative rates. It is challenging to minimize both false negative and positive rates simultaneously. Thus, the maximized Hindex was considered a feasible index to choose an optimal model in this study.\u003c/p\u003e \u003cp\u003eThese state-of-the-art and popular convolutional neural networks (CNN) with transfer learning were regarded as suitable for the classification task between PCa and BPH in this study (Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e). The optimal investigated CNN was EfficientNetV2 in this study.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e"},{"header":"Discussion","content":"\u003cdiv id=\"Sec4\" class=\"Section2\"\u003e \u003ch2\u003eComparisions between presented results and published articles\u003c/h2\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eDeep learning techniques have played an essential role in diagnosing PCa over the past decade [\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e]. In recent articles [\u003cspan additionalcitationids=\"CR29\" citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e], magnetic resonance imaging (MRI) has been commonly used and studied for the segmentation, classification, and detection of PCa in computer-aided diagnosis with artificial intelligence methods. However, only a few deep learning tools have been utilized for classifying PCa with TRUS images. Therefore, the primary objective of this study was to classify PCa with TRUS images and investigate the classification performance of PCa using popular CNNs with transfer learning. The analytical results showed feasible and reasonable classification between PCa and BPH (Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e). The EfficientNetV2, ResANeT, HarDNN, MobileNetV3, DPResNet, CSENeT, and InceptionV3 with transfer learning are useful deep CNN methods for classifying TRUS images. Furthermore, all the accuracy, specificity, sensitivity, PPV (also called Precision), and Hindex values were higher than 0.940, as demonstrated by the investigated CNN methods in this study. The presented methods showcased their ability and efficiency.\u003c/p\u003e \u003cp\u003ePublished results using CNN or SVM are listed in Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e [\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e, \u003cspan additionalcitationids=\"CR43 CR44\" citationid=\"CR42\" class=\"CitationRef\"\u003e42\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e45\u003c/span\u003e]. Feature-based classification between PCa and BPH with ultrasound images has commonly employed SVM classifiers [\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e, \u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e43\u003c/span\u003e, \u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e44\u003c/span\u003e]. However, generating useful features from images remains a challenge. In contrast, CNNs have been applied to B-mode ultrasound to create an efficient end-to-end approach [\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e].\u003c/p\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\u003ethe presented method was compared with the public methods.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"6\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" 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 \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAuthors\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eYear\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eClassification\u0026dagger;\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eMethod\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eModality\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003ePerformance (%)*\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eZhang Q, et. al. [\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e2020\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e47 vs 56\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eSVM\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eB-Mode US\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e87.9/87.0/88.8\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFeng Y, et. al. [\u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e42\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e2018\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e7511 vs 25738\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003ecnn\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eCEUS\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e90.2/82.8/91.5\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eZhiyong L, et. al. [\u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e43\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e2021\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e66 vs 103\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003er-cnn\u0026thinsp;+\u0026thinsp;xception\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eTRUS\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e80.4/92.4/72.8\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHuang X, et. al. [\u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e44\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e2020\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e130 vs 126\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eSVM\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eB-Mode US\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e70.9/70.0/71.7\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eImani F, et. al. [\u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e45\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e2015\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e625 vs 576\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eSVM\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eMRI\u0026thinsp;+\u0026thinsp;RF-Ultrasound\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e80.0/88.0/80.0\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eThis work\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003e2022\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e\u003cb\u003e1380 vs 1530\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u003cb\u003eEfficientNetV2\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u003cb\u003eTRUS\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e\u003cb\u003e99.7/99.8/99.8\u003c/b\u003e\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=\"BlockQuote\"\u003e \u003cp\u003e\u0026dagger; means the sample size between malignant (positive) and benign (negative or BPH). * means accuracy/sensitivity/specificity. CEUS means the contrast-enhanced ultrasound. B-Mode US means brighten mode ultrasound. RF-Ultrasound means radio frequency ultrasound.\u003c/p\u003e \u003cp\u003eFeature-based classification of PCa using ultrasound or contrast-enhanced ultrasound has been conducted by Zhang Q [\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e], Huang X [\u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e44\u003c/span\u003e], and Imani F [\u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e45\u003c/span\u003e]. The highest accuracy, sensitivity, and specificity among these results are 87.9%, 88.0%, and 88.8%. Moreover, deep learning methods have been applied to classify PCa using prostate ultrasound by Feng Y [\u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e42\u003c/span\u003e] and Zhiyong L [\u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e43\u003c/span\u003e]. The highest accuracy, sensitivity, and specificity among these results are 90.2%, 92.4%, and 91.5%. CNN improves the classification performance between PCa and benign prostate ultrasound compared with feature learning methods.\u003c/p\u003e \u003cp\u003eThe accuracy, sensitivity, and specificity achieved by EfficientNetV2 in this study are 99.4%, 99.2%, and 99.5%, respectively (Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e). Transfer learning via CNNs yields satisfactory and acceptable classification results. Indeed, the methods presented in this study utilize state-of-the-art CNN techniques to create a useful model for classifying TRUS images with feasible performance.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec5\" class=\"Section2\"\u003e \u003ch2\u003eUsing the state-of-art pre-trained CNNs\u003c/h2\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eIn this study, the performance of seven state-of-the-art pre-trained CNNs was investigated for classifying TRUS images. The developed architectures of these latest CNN models were based on directed acyclic graph (DAG) networks [\u003cspan additionalcitationids=\"CR36 CR37 CR38 CR39 CR40\" citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e41\u003c/span\u003e]. A DAG network has a more complex architecture, designed from multiple layers. DAGs possess a rich assortment of algorithms needed for non-linear steps in the complicated geometry of multiple CNNs.\u003c/p\u003e \u003cp\u003eA DAG network has useful properties, including reachability, transitive closure, and transitive reduction. As a result, any DAG can quickly optimize and handle multiple layers as input, as well as output from multiple layers. Meanwhile, DAG algorithms are merited for searching the shortest path for designing nodes (or architectural layers). This is one of the reasons why the investigated CNNs in this study could provide high performance classification for TRUS images.\u003c/p\u003e \u003cp\u003eHowever, several parameters used in CNNs can affect the classification performance, including input image size, batch size, number of epochs, learning rates, loss function, and optimizer. Therefore, setting the parameters of a CNN model is a crucial and essential step in building a useful CNN model.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec6\" class=\"Section2\"\u003e \u003ch2\u003eEfficient Detection between BPH and PCa\u003c/h2\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eThe success of these CNN models can be attributed to their ability to adapt to the unique characteristics of TRUS images and the utilization of transfer learning. This approach allowed the models to leverage pre-existing knowledge from large-scale datasets, enabling them to perform better in the classification task. The promising results suggest that the investigated CNNs, particularly the EfficientNetV2, hold great potential as reliable tools for accurately classifying TRUS images and distinguishing between BPH and PCa cases. Furthermore, these findings contribute to the growing body of research on the application of deep learning techniques in medical image analysis, particularly for prostate cancer diagnosis. The use of efficient and accurate deep learning models can significantly improve the clinical decision-making process and ultimately lead to better patient outcomes. By employing state-of-the-art CNNs for the classification of TRUS images, this study has demonstrated the potential of these methods in enhancing the detection of BPH and PCa, which is essential for effective treatment planning and management.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec7\" class=\"Section2\"\u003e \u003ch2\u003eSummary\u003c/h2\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eIn this study, seven state-of-the-art CNN models were investigated for the classification of PCa and BPH using TRUS images. Early detection of prostate cancer (PCa) and benign prostatic hyperplasia (BPH) is crucial for maintaining the health and well-being of aging male populations. This paper highlights the connection between medicine and visual sciences, emphasizing the importance of utilizing advanced techniques to improve diagnostic accuracy.\u003c/p\u003e \u003cp\u003eThe modified TRUS images, with header information removed, prevented interference during the feature extraction process by the CNN models. This removal of header information from TRUS images is considered a crucial step for the successful implementation of CNN methods. With two-fold cross-validation, the accuracy, sensitivity, and specificity achieved were 0.9977, 0.9980, and 0.9980, respectively\u003c/p\u003e \u003cp\u003eThe EfficientNetV2, one of the investigated CNN methods, demonstrated its usefulness and efficiency in classifying TRUS images for differentiating PCa and BPH. Moreover, the combination of deep learning and machine learning methods has the potential to yield high classification accuracy. The presented methods can be applied to assist in the detection of TRUS images, complementing clinical diagnosis made by physicians, and ultimately contributing to better treatment planning and patient outcomes.\u003c/p\u003e \u003cp\u003eTherefore, this study showcases the potential of state-of-the-art CNN models, particularly EfficientNetV2, in improving the classification of TRUS images for PCa and BPH. These findings have implications for enhancing the diagnostic process, leading to more timely and accurate detection, which is essential in ensuring the health and well-being of aging male populations.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003eLimitations amd Future Works\u003c/h2\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eDespite the promising results, there are some limitations to this study that should be addressed and could open avenues for future work. The study used a limited dataset of TRUS images for the classification of PCa and BPH. Expanding the dataset with a larger and more diverse collection of images could further validate the effectiveness of the proposed methods and improve the generalizability of the models. While transfer learning was employed in this study, it may be beneficial to explore other pre-training and fine-tuning strategies to enhance the performance of the CNN models. Additionally, the development of task-specific architectures tailored to the classification of PCa and BPH using TRUS images could lead to even better results.\u003c/p\u003e \u003cp\u003eFuture work could explore the integration of multi-modal data, such as combining TRUS images with MRI or other diagnostic tools, to enhance the detection and classification performance of the models. Developing methods to better understand and visualize the features learned by the CNN models can provide insights into the decision-making process, potentially leading to increased trust and adoption of the models by medical professionals. Future work could investigate the development of real-time classification systems that can be integrated with existing medical imaging devices, providing immediate feedback to clinicians during patient examinations. Investigating the performance of the CNN models in longitudinal studies, where patient data is collected over time, could provide insights into the progression of PCa and BPH, as well as the effectiveness of different treatments.\u003c/p\u003e \u003cp\u003eBy addressing these limitations and exploring future work, the application of CNN models in the classification of PCa and BPH using TRUS images could be further optimized, contributing to more accurate and timely detection and improving patient outcomes.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003c/div\u003e"},{"header":"Methods and Materials","content":"\u003cdiv id=\"Sec10\" class=\"Section2\"\u003e \u003ch2\u003eEthics approval\u003c/h2\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eThis study was conducted after approval by the Institutional Review Board of Kaohsiung Veterans General Hospital (VGHKS IRB; No. VGHKS13-CT6-04). Due to the retrospective nature of the study, informed consent was waived by the VGHKS IRB. This study was conducted with the methods in accordance with relevant guidelines and regulations.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003eThe enrolled samples and research flowchart\u003c/h2\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eThe cases in PCa and BPH were 1,380 and 1,530. The age (years), prostate-specific antigen (PSA (ng/mL)), and TRUS images were collected in this work. The descriptive statistics were shown in Table\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e4\u003c/span\u003e. The mean\u0026thinsp;\u0026plusmn;\u0026thinsp;SD (Standard Deviation, SD) of age between BPH and PCa groups were 64.3\u0026thinsp;\u0026plusmn;\u0026thinsp;9.4 (years) and 68.1\u0026thinsp;\u0026plusmn;\u0026thinsp;9.8 (years). The mean\u0026thinsp;\u0026plusmn;\u0026thinsp;SD of PSA between BPH and PCa were 1.7\u0026thinsp;\u0026plusmn;\u0026thinsp;0.5 (ng/mL) and 28.7\u0026thinsp;\u0026plusmn;\u0026thinsp;45.7 (ng/mL). All of the collected TRUS images were confirmed by pathological biopsy to identify as PCa or BPH.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab4\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 4\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003ethe age and PSA between groups.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"5\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eGroup\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003eAge (years)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"2\" nameend=\"c5\" namest=\"c4\"\u003e \u003cp\u003ePSA (ng/mL)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eMean\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eSTD\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eMean\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eSTD\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBPH(n\u0026thinsp;=\u0026thinsp;1,530)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e64.3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e9.4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e1.7\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.5\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePCa (n\u0026thinsp;=\u0026thinsp;1,380)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e68.1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e9.8\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e28.7\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e45.7\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=\"BlockQuote\"\u003e \u003cp\u003eThe preprocessing of the TRUS images was necessary before training classification models. In order to remove any identifying information (such as patient name, patient ID, hospital name, and other information) from the input images, the modified images were created by setting intensity zeros outside the prostate area (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e). The TRUS images were in the size of 640 \u0026times; 480 with gray-level PNG format. The size and format of the modified images remained the same as the input images. Meanwhile, the extra information might interfere with the classification accuracy using the presented methods. The modified images were created according to the boundary of the dashed line in the input image. The main purpose of creating modified images was to avoid the influences on classification due to the header information around the image (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e right). The header information included patient ID, study date, and scanning parameters (i.e., depth ruler, imaging settings, gray level bar, and measurement information).\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eThe flowchart in this study included input images, preprocessing, pre-training deep learning methods (DL) via transfer learning, validation, and results (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e). In the image processing step, the input images were modified by excluding header information and saved as new images. The investigated popular and latest deep learning approaches included seven deep learning methods, as described in Section 2.2. Next, the investigated models with different parameter settings were validated based on accuracy, specificity, sensitivity, negative predictive value (NPV), positive predictive value (PPV), Kappa, and Hindex values.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003eThe Deep Learning via Transferred Learning\u003c/h2\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eThe DL methods employed the popular convolutional neural networks (CNN), including feature-map based convolutional neural network (HarDNN) [\u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e], InceptionV3 [\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e], MobileNetV3 [\u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e], competitive squeeze and excitation neural network (CSENeT) [\u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e], residual attention neural network (ResANeT) [\u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e], deep pyramidal residual neural network (DPResNet) [\u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e40\u003c/span\u003e], and EfficientNetV2 [\u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e41\u003c/span\u003e] via transfer learning (Table\u0026nbsp;\u003cspan refid=\"Tab5\" class=\"InternalRef\"\u003e5\u003c/span\u003e). The investigated CNN methods were regarded as the best deep CNN architectures for classification tasks. Meanwhile, the classification layer and fully connected layer were modified and fixed for two classes (i.e., the original designed number of classes was 1,000). Additionally, a batch size of 5 was investigated for the optimal accuracy of the presented methods. The two-fold cross-validation was designed to evaluate the classification performance for the presented methods.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab5\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 5\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eThe merits of investigated CNNs in recently.\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=\"char\" char=\".\" 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\u003eCNN\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAuthors\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eYear\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eMerits\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHarDNN\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eMahmoud et. al. [\u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e2020\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eThe HarDNN was accurately estimated a feature map in CNN. HarDNN was demonstrated to improve computational times by 10x faster than other CNNs.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eInceptionV3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eKin N. [\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e2019\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eThe InceptionV3 was proposed to prompt the accuracy of recognition by performing a fine-tuned parameter in the layers of CNN with transferred learning pre-trained model. The InceptionV3 was trained and validated on a cooking dataset. The InceptionV3 was provide a potential solution to the recognition problem.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMobileNetV3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eChu et. al. [\u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e2019\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eMobileNet was used the depth-wise separable convolutions to create a small weight CNN. The global hyper-parameters of MobileNet that efficiently balance between latent parameters and classification accuracy by evaluating ImageNet classification. Meanwhile, MobileNet had demonstrated the across a wide range of applications including object detection, fine-grain classification, face attributes and large-scale geo-localization.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCSENeT\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eHu et. al. [\u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e2018\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eThe competitive squeeze-excitation residual network (CSENet) was built for to determined and to expand the meaning of channel relationship in residual layers. The CSENet was proven as good as the popular CNN.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eResANeT\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eWang et. al. [\u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e2017\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eThe residual attention network (ResANeT) was built according stacking attention modules with generating attention-aware features. The ResANeT not only be used to train deep residual attention network, but could be easily adjusted to hundreds of layers.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDPResNet\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eHan et.al. [\u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e40\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e2016\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eThe deep pyramidal residual neural network (DPResNet) was designed to ensure effective performance by increasing the diversity of high-level features in an image. The DPResNet had been proven to improve generalization ability to prompt the classification accuracy.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eEfficientNetV2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTan et.al. [\u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e41\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e2021\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eThe EfficientNetV2 was a new CNN that trains faster and efficiency parameters than other CNN models. The combination of neural architecture searching and scaling with adding optimization. The EfficientNetV2 significantly outperforms previous models on ImageNet and CIFAR/Cars/Flowers 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 \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003eEvaluated Perofrmance of Presented Methods\u003c/h2\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eThe performance of the presented models was evaluated using the testing set. The testing set consisted of 50% randomly sampled data from each group, with two-fold cross-validation. The testing performance of the presented methods is typically assessed using popular indices. A confusion matrix is often employed in the literature to evaluate the suitability of different models, including their sensitivity, specificity, accuracy, positive predictive value (PPV), negative predictive value (NPV), Kappa value, and Hindex index (or Youden's index).\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003c/div\u003e"},{"header":"Abbreviations","content":"\u003cp\u003eThe abbreviations used in this text are listed with their full names below.\u003c/p\u003e\n\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\" width=\"100%\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd width=\"36.36363636363637%\"\u003e\n \u003cp\u003e\u003cstrong\u003eAbbreviation\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"63.63636363636363%\"\u003e\n \u003cp\u003e\u003cstrong\u003eFull Name\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"36.36363636363637%\"\u003e\n \u003cp\u003eAI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"63.63636363636363%\"\u003e\n \u003cp\u003eArtificial intelligent\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"36.36363636363637%\"\u003e\n \u003cp\u003eBPH\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"63.63636363636363%\"\u003e\n \u003cp\u003eBenign prostatic hyperplasia\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"36.36363636363637%\"\u003e\n \u003cp\u003eCNN\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"63.63636363636363%\"\u003e\n \u003cp\u003eConvolutional neuron networks\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"36.36363636363637%\"\u003e\n \u003cp\u003eCSENeT\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"63.63636363636363%\"\u003e\n \u003cp\u003eCompetitive squeeze and excitation neural network\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"36.36363636363637%\"\u003e\n \u003cp\u003eCT\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"63.63636363636363%\"\u003e\n \u003cp\u003eComputed tomography\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"36.36363636363637%\"\u003e\n \u003cp\u003eDAG\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"63.63636363636363%\"\u003e\n \u003cp\u003eDirected acyclic graph\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"36.36363636363637%\"\u003e\n \u003cp\u003eDL\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"63.63636363636363%\"\u003e\n \u003cp\u003eDeep learning\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"36.36363636363637%\"\u003e\n \u003cp\u003eDPResNet\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"63.63636363636363%\"\u003e\n \u003cp\u003eDeep pyramidal residual neural network\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"36.36363636363637%\"\u003e\n \u003cp\u003eDR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"63.63636363636363%\"\u003e\n \u003cp\u003eDigital radiography\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"36.36363636363637%\"\u003e\n \u003cp\u003eHarDNN\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"63.63636363636363%\"\u003e\n \u003cp\u003eFeature-map based convolutional neural network\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"36.36363636363637%\"\u003e\n \u003cp\u003eHindex\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"63.63636363636363%\"\u003e\n \u003cp\u003eYouden\u0026apos;s index\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"36.36363636363637%\"\u003e\n \u003cp\u003eMRI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"63.63636363636363%\"\u003e\n \u003cp\u003eMagnetic resonance imaging\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"36.36363636363637%\"\u003e\n \u003cp\u003eNPV\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"63.63636363636363%\"\u003e\n \u003cp\u003eNegative predictive value\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"36.36363636363637%\"\u003e\n \u003cp\u003epatient ID\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"63.63636363636363%\"\u003e\n \u003cp\u003ePatient identification\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"36.36363636363637%\"\u003e\n \u003cp\u003ePCa\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"63.63636363636363%\"\u003e\n \u003cp\u003eProstate cancer\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"36.36363636363637%\"\u003e\n \u003cp\u003ePET\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"63.63636363636363%\"\u003e\n \u003cp\u003ePositron emission tomography\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"36.36363636363637%\"\u003e\n \u003cp\u003ePPV\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"63.63636363636363%\"\u003e\n \u003cp\u003ePositive predictive value or Precision\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"36.36363636363637%\"\u003e\n \u003cp\u003ePSA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"63.63636363636363%\"\u003e\n \u003cp\u003eProstate-specific antigen\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"36.36363636363637%\"\u003e\n \u003cp\u003eResANeT\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"63.63636363636363%\"\u003e\n \u003cp\u003eResidual attention neural network\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"36.36363636363637%\"\u003e\n \u003cp\u003eSTD\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"63.63636363636363%\"\u003e\n \u003cp\u003eStandard Deviation\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"36.36363636363637%\"\u003e\n \u003cp\u003eTRUS\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"63.63636363636363%\"\u003e\n \u003cp\u003eTransrectal ultrasound\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eAuthor Contributions:\u0026nbsp;\u003c/strong\u003eConceptualization, T.L.H. and T.B.C.; methodology, N.H.L., Y.H.H, and T.B.C; software, W.H.T.; validation, N.H.L. and T.B.C.; formal analysis, T.L.H.; investigation, T.L.H.; resources, Y.H.H; data curation T.L.H. and K.Y.L; writing\u0026mdash;original draft preparation, T.L.H., T.B.C. and N.H.L.; writing\u0026mdash;review and editing, T.B.C. and N.H.L.; visualization, L.R.Y. and K.YL; supervision, T.B.C.; project administration T.B.C. All authors have read and agreed to the published version of the manuscript.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding:\u003c/strong\u003e The authors would like to thank the National Science and Technology Council in Taiwan for financially supporting this research under contract NSTC 111-2118-M-214-001 and 110-2118-M-214-001.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting interests:\u0026nbsp;\u003c/strong\u003eThe authors declare no conflict of interest.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eEthics approval and consent to participate Consent for publication:\u0026nbsp;\u003c/strong\u003eThis study was conducted after approval by the Institutional Review Board of Kaohsiung Veterans General Hospital (VGHKS IRB; No. VGHKS13-CT6-04). Due to the retrospective nature of the study, informed consent was waived by the VGHKS IRB. This study was conducted with the methods in accordance with relevant guidelines and regulations.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAvailability of data and material:\u0026nbsp;\u003c/strong\u003eThe data presented in this study are available upon request from the corresponding author. The data are not publicly available due to restrictions, e.g., privacy and ethical concerns.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAcknowledgments:\u003c/strong\u003e This research was partially supported by grants from the National Science and Technology Council, R.O.C (NSTC110-2118-M-214-001, and NSTC 111-2118-M-214-001).\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eMorgan N, Van Gerven A, Smolders A, de Faria Vasconcelos K, Willems H, Jacobs R. Convolutional neural network for automatic maxillary sinus segmentation on cone-beam computed tomographic images. Sci Rep. 2022 May 7;12(1):7523. doi:10.1038/s41598-022-11483-3\u003c/li\u003e\n\u003cli\u003eDeng Y, Wang L, Zhao C, Tang S, Cheng X, Deng HW, Zhou W. A deep learning-based approach to automatic proximal femur segmentation in quantitative CT images. Med Biol Eng Comput. 2022 May;60(5):1417-1429. doi:10.1007/s11517-022-02529-9.\u003c/li\u003e\n\u003cli\u003eHung KF, Ai QYH, King AD, Bornstein MM, Wong LM, Leung YY. Automatic detection and segmentation of morphological changes of the maxillary sinus mucosa on cone-beam computed tomography images using a three-dimensional convolutional neural network. Clin Oral Investig. 2022 May;26(5):3987-3998. doi:10.1007/s00784-021-04365-x.\u003c/li\u003e\n\u003cli\u003eValjarević, A.; Morar, C.; Živković, J.; Niemets, L.; Kićović, D.; Golijanin, J.; Gocić, M.; Bursać, N.M.; Stričević, L.; Žiberna, I.; Bačević, N.; Milevski, I.; Durlević, U.; Lukić, T. Long Term Monitoring and Connection between Topography and Cloud Cover Distribution in Serbia. Atmosphere 2021, 12, 964. Doi: 10.3390/atmos12080964. \u003c/li\u003e\n\u003cli\u003eBerhane H, Scott MB, Barker AJ, McCarthy P, Avery R, Allen B, Malaisrie C, Robinson JD, Rigsby CK, Markl M. Deep learning-based velocity antialiasing of 4D-flow MRI. Magn Reson Med. 2022 Jul;88(1):449-463. doi: 10.1002/mrm.29205.\u003c/li\u003e\n\u003cli\u003eLuciw NJ, Shirzadi Z, Black SE, Goubran M, MacIntosh BJ. Automated generation of cerebral blood flow and arterial transit time maps from multiple delay arterial spin-labeled MRI. Magn Reson Med. 2022 Jul;88(1):406-417. doi:10.1002/mrm.29193.\u003c/li\u003e\n\u003cli\u003eArabahmadi M, Farahbakhsh R, Rezazadeh J. Deep Learning for Smart Healthcare-A Survey on Brain Tumor Detection from Medical Imaging. Sensors(Basel). 2022 Mar 2;22(5):1960. doi: 10.3390/s22051960.\u003c/li\u003e\n\u003cli\u003eAggarwal P, Mishra NK, Fatimah B, Singh P, Gupta A, Joshi SD. COVID-19 image classification using deep learning: Advances, challenges and opportunities. Comput Biol Med. 2022 May;144:105350. doi:10.1016/j.compbiomed.2022.105350.\u003c/li\u003e\n\u003cli\u003eAggarwal P, Mishra NK, Fatimah B, Singh P, Gupta A, Joshi SD. COVID-19 image classification using deep learning: Advances, challenges and opportunities. Comput Biol Med. 2022 May;144:105350. doi: 10.1016/j.compbiomed.2022.105350.\u003c/li\u003e\n\u003cli\u003eMulrenan C, Rhode K, Fischer BM. A Literature Review on the Use of Artificial Intelligence for the Diagnosis of COVID-19 on CT and Chest X-ray. Diagnostics (Basel). 2022 Mar 31;12(4):869. doi: 10.3390/diagnostics12040869.\u003c/li\u003e\n\u003cli\u003eMinoshima S, Cross D. Application of artificial intelligence in brain molecular imaging. Ann Nucl Med. 2022 Feb;36(2):103-110. doi:10.1007/s12149-021-01697-2.\u003c/li\u003e\n\u003cli\u003eXue S, Guo R, Bohn KP, Matzke J, Viscione M, Alberts I, Meng H, Sun C, Zhang M, Zhang M, Sznitman R, El Fakhri G, Rominger A, Li B, Shi K. A cross-scanner and cross-tracer deep learning method for the recovery of standard-dose imaging quality from low-dose PET. Eur J Nucl Med Mol Imaging. 2022 May;49(6):1843-1856. doi:10.1007/s00259-021-05644-1.\u003c/li\u003e\n\u003cli\u003eDal Toso L, Chalampalakis Z, Buvat I, Comtat C, Cook G, Goh V, Schnabel JA, Marsden PK. Improved 3D tumour definition and quantification of uptake in simulated lung tumours using deep learning. Phys Med Biol. 2022 Apr 27;67(9). doi:10.1088/1361-6560/ac65d6.\u003c/li\u003e\n\u003cli\u003eMamistvalov A, Amar A, Kessler N, Eldar YC. Deep-Learning Based Adaptive Ultrasound Imaging From Sub-Nyquist Channel Data. IEEE Trans Ultrason Ferroelectr Freq Control. 2022 May;69(5):1638-1648. doi:10.1109/TUFFC.2022.3160859.\u003c/li\u003e\n\u003cli\u003eSato M, Kobayashi T, Soroida Y, Tanaka T, Nakatsuka T, Nakagawa H, Nakamura A, Kurihara M, Endo M, Hikita H, Sato M, Gotoh H, Iwai T, Tateishi R, Koike K, Yatomi Y. Development of novel deep multimodal representation learning-based\u003c/li\u003e\n\u003cli\u003emodel for the differentiation of liver tumors on B-mode ultrasound images. J Gastroenterol Hepatol. 2022 Apr;37(4):678-684. doi: 10.1111/jgh.15763.\u003c/li\u003e\n\u003cli\u003eSong Y, Zheng J, Lei L, Ni Z, Zhao B, Hu Y. CT2US: Cross-modal transfer learning for kidney segmentation in ultrasound images with synthesized data. Ultrasonics. 2022 May;122:106706. doi: 10.1016/j.ultras.2022.106706.\u003c/li\u003e\n\u003cli\u003eHuang X, Chen M, Liu P, Du Y. Texture Feature-Based Classification on Transrectal Ultrasound Image for Prostatic Cancer Detection. Comput Math Methods Med. 2020 Oct 6;2020:7359375. doi: 10.1155/2020/7359375.\u003c/li\u003e\n\u003cli\u003eAleksandar V, Tatjana D, Vladica S, Radomir I, Bojana J, GIS numerical and remote sensing analyses of forest changes in the Toplica region for the period of 1953\u0026ndash;2013. Applied Geography. 2018;92:131-139. doi:10.1016/j.apgeog.2018.01.016.\u003c/li\u003e\n\u003cli\u003eAhmad, J., Farman, H., Jan, Z. (2019). Deep Learning Methods and Applications. In: Deep Learning: Convergence to Big Data Analytics. Springer Briefs in Computer Science. Springer, Singapore. Doi: 10.1007/978-981-13-3459-7_3.\u003c/li\u003e\n\u003cli\u003eFaure Walker N, Gill B, Olsburgh J, Gillatt D, Yap T, Michala L, Taylor C, Wood H, Wood D. Age-related urologic problems in the complex urologic patient. World J Urol. 2021 Apr;39(4):1037-1044. doi:10.1007/s00345-020-03111-4.\u003c/li\u003e\n\u003cli\u003eSedghi A, Pesteie M, Javadi G, Azizi S, Yan P, Kwak JT, Xu S, Turkbey B, Choyke P, Pinto P, Wood B, Rohling R, Abolmaesumi P, Mousavi P. Deep neural maps for unsupervised visualization of high-grade cancer in prostate biopsies. Int J Comput Assist Radiol Surg. 2019 Jun;14(6):1009-1016. doi:10.1007/s11548-019-01950-0.\u003c/li\u003e\n\u003cli\u003eWasserman NF, Niendorf E, Spilseth B. Precision and accuracy of magnetic resonance imaging for lobar classification of benign prostatic hyperplasia. Abdom Radiol (NY). 2019 Jul;44(7):2535-2544. doi:10.1007/s00261-019-01970-z.\u003c/li\u003e\n\u003cli\u003eWang X, Yang W, Weinreb J, Han J, Li Q, Kong X, Yan Y, Ke Z, Luo B, Liu T, Wang L. Searching for prostate cancer by fully automated magnetic resonance imaging classification: deep learning versus non-deep learning. Sci Rep. 2017 Nov 13;7(1):15415. doi: 10.1038/s41598-017-15720-y.\u003c/li\u003e\n\u003cli\u003eSong Y, Zhang YD, Yan X, Liu H, Zhou M, Hu B, Yang G. Computer-aided diagnosis of prostate cancer using a deep convolutional neural network from multiparametric MRI. J Magn Reson Imaging. 2018 Dec;48(6):1570-1577. doi:10.1002/jmri.26047.\u003c/li\u003e\n\u003cli\u003eAldoj N, Lukas S, Dewey M, Penzkofer T. Semi-automatic classification of prostate cancer on multi-parametric MR imaging using a multi-channel 3D convolutional neural network. Eur Radiol. 2020 Feb;30(2):1243-1253. doi:10.1007/s00330-019-06417-z.\u003c/li\u003e\n\u003cli\u003eYuan Y, Qin W, Buyyounouski M, Ibragimov B, Hancock S, Han B, Xing L. Prostate cancer classification with multiparametric MRI transfer learning model. Med Phys. 2019 Feb;46(2):756-765. doi: 10.1002/mp.13367.\u003c/li\u003e\n\u003cli\u003eWildeboer RR, van Sloun RJG, Wijkstra H, Mischi M. Artificial intelligence in multiparametric prostate cancer imaging with focus on deep-learning methods. Comput Methods Programs Biomed. 2020 Jun;189:105316. doi:10.1016/j.cmpb.2020.105316.\u003c/li\u003e\n\u003cli\u003eTătaru OS, Vartolomei MD, Rassweiler JJ, Virgil O, Lucarelli G, Porpiglia F, Amparore D, Manfredi M, Carrieri G, Falagario U, Terracciano D, de Cobelli O, Busetto GM, Del Giudice F, Ferro M. Artificial Intelligence and Machine Learning in Prostate Cancer Patient Management-Current Trends and Future Perspectives. Diagnostics (Basel). 2021 Feb 20;11(2):354.\u003c/li\u003e\n\u003cli\u003eSyer T, Mehta P, Antonelli M, Mallett S, Atkinson D, Ourselin S, Punwani S. Artificial Intelligence Compared to Radiologists for the Initial Diagnosis of Prostate Cancer on Magnetic Resonance Imaging: A Systematic Review and Recommendations for Future Studies. Cancers (Basel). 2021 Jul 1;13(13):3318. doi: 10.3390/cancers13133318.\u003c/li\u003e\n\u003cli\u003eZhang Q, Xiong J, Cai Y, Shi J, Xu S, Zhang B. Multimodal feature learning and fusion on B-mode ultrasonography and sonoelastography using point-wise gated deep networks for prostate cancer diagnosis. Biomed Tech (Berl). 2020 Jan 28;65(1):87-98. doi: 10.1515/bmt-2018-0136.\u003c/li\u003e\n\u003cli\u003eWildeboer RR, Mannaerts CK, van Sloun RJG, Bud\u0026auml;us L, Tilki D, Wijkstra H, Salomon G, Mischi M. Automated multiparametric localization of prostate cancer based on B-mode, shear-wave elastography, and contrast-enhanced ultrasound radiomics. Eur Radiol. 2020 Feb;30(2):806-815. doi: 10.1007/s00330-019-06436-w.\u003c/li\u003e\n\u003cli\u003eShah M, Naik N, Somani BK, Hameed BMZ. Artificial intelligence (AI) in urology-Current use and future directions: An iTRUE study. Turk J Urol. 2020 Nov;46(Supp. 1):S27-S39. doi: 10.5152/tud.2020.20117.\u003c/li\u003e\n\u003cli\u003eChiu PK, Shen X, Wang G, Ho CL, Leung CH, Ng CF, Choi KS, Teoh JY. Enhancement of prostate cancer diagnosis by machine learning techniques: an algorithm development and validation study. Prostate Cancer Prostatic Dis. 2021 Jul 15. doi: 10.1038/s41391-021-00429-x.\u003c/li\u003e\n\u003cli\u003eMahmoud A, Hari S, Fletcher CW, Adve SV, Sakr C, Shanbhag N, Molchanov P, Sullivan MB, Tsai T, Keckler SW. HarDNN: Feature Map Vulnerability Evaluation in CNNs. 2020 Feb; arXiv:2002.09786. doi:10.48550/arXiv.2002.09786,\u003c/li\u003e\n\u003cli\u003eKin N. Tuned Inception V3 for Recognizing States of Cooking Ingredients. 2019 May; arXiv:1905.03715. doi: 10.48550/arXiv.1905.03715.\u003c/li\u003e\n\u003cli\u003eChu X, Zhang B, Xu R. MoGA: Searching Beyond MobileNetV3. 2019 Aug; arXiv:1908.01314. doi: 10.48550/arXiv.1908.01314.\u003c/li\u003e\n\u003cli\u003eHu Y, Wen G, Luo M, Dai D, Ma J, Yu Z. Competitive Inner-Imaging Squeeze and Excitation for Residual Network. 2018 Jul; arXiv:1807.08920. doi:10.48550/arXiv.1807.08920.\u003c/li\u003e\n\u003cli\u003eWang F, Jiang M, Qian C, Yang S, Li C, Zhang H, Wang X, Tang X. Residual Attention Network for Image Classification. 2017 Apr; arXiv:1704.06904. doi:10.48550/arXiv.1704.06904.\u003c/li\u003e\n\u003cli\u003eHan D, Kim J, Kim J. Deep Pyramidal Residual Networks. 2019 Oct; arXiv:1610.02915. doi:10.48550/arXiv.1610.02915.\u003c/li\u003e\n\u003cli\u003eTan M, Le QV. EfficientNetV2: Smaller Models and Faster Training. 2021 Apr; arXiv:2104.00298. doi:10.48550/arXiv.2104.00298\u003c/li\u003e\n\u003cli\u003eFeng Y, Yang F, Zhou X, Guo Y, Tang F, Ren F, Guo J, Ji S. A Deep Learning Approach for Targeted Contrast-Enhanced Ultrasound Based Prostate Cancer Detection. IEEE/ACM Trans Comput Biol Bioinform. 2019 Nov-Dec;16(6):1794-1801. doi: 10.1109/TCBB.2018.2835444.\u003c/li\u003e\n\u003cli\u003eZhiyong L, Chuan Y, Jun H, Shaopeng L, Yumin Z. Xu L. Deep learning framework based on integration of S-Mask R-CNN and Inception-v3 for ultrasound image-aided diagnosis of prostate cancer. Future Generation Computer Systems. 2021 Jan; 114:258-367. doi:10.1016/j.future.2020.08.015.\u003c/li\u003e\n\u003cli\u003eHuang X, Chen M, Liu P, Du Y. Texture Feature-Based Classification on Transrectal Ultrasound Image for Prostatic Cancer Detection. Comput Math Methods Med. 2020 Oct 6;2020:7359375. doi: 10.1155/2020/7359375. \u003c/li\u003e\n\u003cli\u003eImani F, Abolmaesumi P, Gibson E, Khojaste A, Gaed M, Moussa M, Gomez JA, Romagnoli C, Leveridge M, Chang S, Siemens DR, Fenster A, Ward AD, Mousavi P. Computer-Aided Prostate Cancer Detection Using Ultrasound RF Time Series: In Vivo Feasibility Study. IEEE Trans Med Imaging. 2015 Nov;34(11):2248-57. doi:10.1109/TMI.2015.2427739.\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"Prostate Cancer, TRUS, BPH, CNN, Transfer Learning","lastPublishedDoi":"10.21203/rs.3.rs-2853191/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-2853191/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003e\u003cb\u003ePurpose\u003c/b\u003e\u003c/p\u003e \u003cp\u003eEarly detection of prostate cancer (PCa) and benign prostatic hyperplasia (BPH) is crucial for maintaining the health and well-being of aging male populations. This study aims to evaluate the performance of transfer learning with convolutional neural networks (CNNs) for efficient classification of PCa and BPH in transrectal ultrasound (TRUS) images.\u003c/p\u003e\u003cp\u003e\u003cb\u003eMethods\u003c/b\u003e\u003c/p\u003e \u003cp\u003eA retrospective experimental design was employed in this study, with 1,380 TRUS images for PCa and 1,530 for BPH. Seven state-of-the-art deep learning (DL) methods were employed as classifiers with transfer learning applied to popular CNN architectures. Performance indices, including sensitivity, specificity, accuracy, positive predictive value (PPV), negative predictive value (NPV), Kappa value, and Hindex (Youden's index), were used to assess the feasibility and efficacy of the CNN methods.\u003c/p\u003e\u003cp\u003e\u003cb\u003eResults\u003c/b\u003e\u003c/p\u003e \u003cp\u003eThe CNN methods with transfer learning demonstrated a high classification performance for TRUS images, with all accuracy, specificity, sensitivity, PPV, NPV, Kappa, and Hindex values surpassing 0.9400. The optimal accuracy, sensitivity, and specificity reached 0.9987, 0.9980, and 0.9980, respectively, as evaluated using two-fold cross-validation.\u003c/p\u003e\u003cp\u003e\u003cb\u003eConclusion\u003c/b\u003e\u003c/p\u003e \u003cp\u003eThe investigated CNN methods with transfer learning showcased their efficiency and ability for the classification of PCa and BPH in TRUS images. Notably, the EfficientNetV2 with transfer learning displayed a high degree of effectiveness in distinguishing between PCa and BPH, making it a promising tool for future diagnostic applications.\u003c/p\u003e","manuscriptTitle":"Transfer Learning with CNNs for Efficient Prostate Cancer and BPH Detection in Transrectal Ultrasound Images","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2023-05-12 14:50:47","doi":"10.21203/rs.3.rs-2853191/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Major revision","date":"2023-08-17T05:24:32+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2023-08-16T08:04:00+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2023-08-09T04:12:39+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2023-07-28T22:31:53+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"d8fb6a05-4958-4dff-966f-c536b4631ad4","date":"2023-07-19T21:56:20+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"616f4f28-7c02-4f95-bb07-ba0a6c7e5640","date":"2023-07-19T04:17:55+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2023-07-18T18:34:23+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2023-07-18T18:33:35+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2023-05-10T11:19:20+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2023-05-10T11:16:16+00:00","index":"","fulltext":""},{"type":"submitted","content":"Scientific Reports","date":"2023-04-24T06:35:27+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"a60a5170-0e4a-4f91-aa28-83c50d2cae4b","owner":[],"postedDate":"May 12th, 2023","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[],"tags":[],"updatedAt":"2023-12-11T15:05:06+00:00","versionOfRecord":{"articleIdentity":"rs-2853191","link":"https://doi.org/10.1038/s41598-023-49159-1","journal":{"identity":"scientific-reports","isVorOnly":false,"title":"Scientific Reports"},"publishedOn":"2023-12-09 15:01:42","publishedOnDateReadable":"December 9th, 2023"},"versionCreatedAt":"2023-05-12 14:50:47","video":"","vorDoi":"10.1038/s41598-023-49159-1","vorDoiUrl":"https://doi.org/10.1038/s41598-023-49159-1","workflowStages":[]},"version":"v1","identity":"rs-2853191","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-2853191","identity":"rs-2853191","version":["v1"]},"buildId":"7rjqhiLT3MXkJMwkYKINL","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}