We modified the number of neurons of the last fully-connected layer as 2 for binary classification and 8 for multi-class classification. Vinh, N. X., Epps, J., and Bailey, J. The value of TP in the equations above is the number of images correctly recognized as malignant tumor in the testing subset. The second ensemble consists of a Multi-Layer Perceptron ensemble which focuses on rejected samples from the first ensemble. The very simple and fast, typical clustering algorithm K-means is adopted in this paper to perform this clustering analysis. Alom MZ, Yakopcic C, Nasrin MS, Taha TM, Asari VK. However, multi-class classification is more significant than binary classification for providing accurate treatment and prognosis for breast cancer patients. The average s(i) of all samples in a cluster is a measure of how tightly grouped all the samples in the cluster are. There are 2 encode layers with neuron sizes of 500 and 2, respectively, and there are 2 corresponding decode layers to reconstruct the original input. However, traditional feature extraction methods can only extract some low-level features of images, and prior knowledge is necessary to select useful features, which can be greatly affected by humans. Sp in (3) expresses the ratio of the recognized benign tumor images to all benign tumor images. It is usually set as α = 0.05. Comput. Detection of breast cancer on digital histopathology images: present status and future possibilities. The modified Inception_V3 network structure is similar, so it is omitted. The best results were also obtained using the extended datasets. Keywords: Deep learning techniques can extract high-level abstract features from images automatically. MacQueen, J. So, we output the confusion matrix of multi-class classification for further analysis. Besides using filters of different sizes in the network, the deterioration caused by increasing layers can also be solved by jumping layers as allowed by the use of residual connections. arXiv:180306626. Deep learning techniques can extract high-level abstract features from images automatically. 10.1016/j.procs.2016.04.224 Dermatologist-level classification of skin cancer with deep neural networks. Med. Computer-aided diagnosis (CAD) approaches for automatic diagnoses improve efficiency by allowing pathologists to focus on more difficult diagnosis cases. Then, we froze all of the parameters before the last layer of the networks. doi: 10.7717/peerj.8668. doi: 10.1001/jama.2016.17216, Han, Z., Wei, B., Zheng, Y., Yin, Y., Li, K., and Li, S. (2017). Get the latest public health information from CDC: https://www.coronavirus.gov, Get the latest research information from NIH: https://www.nih.gov/coronavirus, Find NCBI SARS-CoV-2 literature, sequence, and clinical content: https://www.ncbi.nlm.nih.gov/sars-cov-2/. Table 5. Biometrics 33, 159–174. *Correspondence: Juanying Xie, xiejuany@snnu.edu.cn Chaoyang Zhang, chaoyang.zhang@usm.edu, Front. Szegedy, C., Ioffe, S., Vanhoucke, V., and Alemi, A. First, we adapted Inception_V3 and Inception_ResNet_V2 architectures to the binary and multi-class issues of breast cancer histopathological image classification … These classifiers were tested on a set of 737 microscopic images of fine needle biopsies obtained from 67 patients, which contained 25 benign (275 images) and 42 malignant (462 images) cases. In this paper, histopathological images are used as a dataset from Kaggle. (2016a) published a breast cancer dataset called BreaKHis in 2016.  |  Bayramoglu, N., Kannala, J., and Heikkilä, J. The measurement of observer agreement for categorical data. Therefore, we adopt two deep convolutional neural networks, specifically Inception_V3 and Inception_Resnet_V2, to study the diagnosis of breast cancer in the BreaKHis dataset via transfer learning techniques. The network structures, (A) Inception_V3, (B) Inception_ResNet_V2. We present a Multi-Resolution Convolutional Network (MR-CN) with Plurality Voting (MR-CN-PV) model for automated NAS. Kowal, M., Filipczuk, P., Obuchowicz, A., Korbicz, J., and Monczak, R. (2013). On the other hand, by combining deep learning with clustering and utilizing the dimension-reduction functionality of the autoencoder network (Hinton and Salakhutdinov, 2006), we propose a new autoencoder network structure to apply non-linear transformations to features in histopathological images of breast cancer extracted by the Inception_ResNet_V2 network. Counting mitoses in breast cancer histopathological images is a tedious and time consuming task, but it is very important in grading cancer, therefore to help pathologist, an automated system is proposed. This is especially true for the results on the augmented datasets where Se>98%, Sp>99%, PPV>99%, and DOR>100. Sci. Silhouettes: a graphical aid to the interpretation and validation of cluster analysis. GK201701006 and GK201806013. FN is the number of images incorrectly recognized as benign tumor in the testing subset. Representation learning: A review and new perspectives, in IEEE Transactions on Pattern Analysis and Machine Intelligence 35, 1798–1828. (eds) (2016). Further, we have proposed a framework based … (eds) (2015). Here, MI(U, V) denotes the mutual information between two partitions U and V, and E{MI(U, V)} represents the expected mutual information between the original partition U and the clustering V. H(U), H(V) are the entropy of the original partition U and the clustering V, respectively. Comput Methods Programs Biomed. Machine Vision Appl. Since the operation of Whole-Slide Imaging is complex and expensive, many studies based on this technique use small datasets and achieve poor generalization performance. FP is the number of images that were incorrectly recognized as malignant tumor in the testing subset. Eng. To realize the development of a system for diagnosing breast cancer using multi-class classification on BreaKHis, Han et al. The 1,536-dimension features are extracted by using Inception_ResNet_V2 to process histopathological images of breast cancer, and the K-means clustering algorithm is adopted to group these images into proper clusters. Veta, M., Pluim, J. P., van Diest, P. J., and Viergever, M. A. The trends of the other magnification factor datasets are similar. Therefore, the diagnosis of breast cancer has become very important. Our experimental results of the supervised histopathological image classification of breast cancer and the comparison to the results from other studies demonstrate that Inception_V3 and Inception_ResNet_V2 based histopathological image classification of breast cancer is superior to the existing methods. They were able to achieve an accuracy of 98.7–96.4% for binary classification and multi-class classification, respectively. The highest average accuracy achieved … J. Science 313, 504–507. The automatic diagnosis of breast cancer by analyzing histopathological images plays a significant role for patients and their prognosis. No use, distribution or reproduction is permitted which does not comply with these terms. histopathological images contain sufficient phenotypic information, they play an indispensable role in the di-agnosis and treatment of breast cancers. This subsection will compare the clustering results of IRV2+AE+Kmeans and IRV2+Kmeans in terms of external criteria, including ACC, ARI, AMI, and the internal metric SSE. According to the description of the histopathological image dataset of breast cancer, the benign and malignant tumors can be classified into four different subclasses, respectively. doi: 10.1016/j.protcy.2016.05.165, Moraga-Serrano, P. E. (2018). Hodneland E, Dybvik JA, Wagner-Larsen KS, Šoltészová V, Munthe-Kaas AZ, Fasmer KE, Krakstad C, Lundervold A, Lundervold AS, Salvesen Ø, Erickson BJ, Haldorsen I. Sci Rep. 2021 Jan 8;11(1):179. doi: 10.1038/s41598-020-80068-9. Stenkvist, B., Westman-Naeser, S., Holmquist, J., Nordin, B., Bengtsson, E., Vegelius, J., et al. Unlocked 8, 74–79. In 2016, a performance comparison was conducted by Asri et al. Being able to automate the detection of metastasised cancer in pathological scans with machine learning and deep neural networks is an area of medical imaging and diagnostics with promising potential for clinical usefulness. (2008). USA.gov. Therefore, to determine whether the predictions are due to chance, we calculate the p-values for AUC and Kappa and compare the p-value to the significance level α. Histopathological images are mainly used in diagnosis purpose .This paper mainly explains the techniques for detection of breast cancer applying both image processing and deep learning techniques. The evaluation criteria of clustering results comprise internal and external metrics. Zhang, Y., Zhang, B., Coenen, F., and Lu, W. (2013). The automatic diagnosis of breast cancer by analyzing histopathological images plays a significant role for patients and their prognosis. Therefore, we adopt clustering techniques to study the histopathological images of breast cancer. Then, we can retrain the last defined fully-connected layer of the model using only a relatively small amount of data to achieve good results for our target task. doi: 10.1109/TMI.2013.2275151, George, Y. M., Zayed, H. H., Roushdy, M. I., and Elbagoury, B. M. (2014). Int J Comput Assist Radiol Surg. The statistical significance between pairs of algorithms is displayed in the lower triangle using “*.”. “Breast cancer histopathological image classification using convolutional neural networks,” in 2016 International Joint Conference on Neural Networks (IJCNN). The network structures of our proposed autoencoder and its combination with Inception_ResNet_V2. After doing this, the sample number of each subclass was approximately the same. of Breast cancer required new deep learning and transfer learning techniques. Inception-v4, Inception-Resnet and the Impact of Residual Connections on Learning. 2018 Feb;13(2):179-191. doi: 10.1007/s11548-017-1663-9. The values of Kappa in Table 4 show that our models are perfect when applied to augmented datasets for multi-class classification. We demonstrate that our experimental results are superior to the ones available in other studies that we have found, and that the Inception_ResNet_V2 network is more suitable for performing analysis of the histopathological images of breast cancer than the Inception_V3 network. Table 1. 1 Breast Cancer Histopathological Image Classification: A Deep Learning Approach Mehdi Habibzadeh Motlagh1, Mahboobeh Jannesari2, HamidReza Aboulkheyr1, Pegah Khosravi3, Olivier Elemento 3,*, Mehdi Totonchi1,2,*, and Iman Hajirasouliha 1Department of Stem Cells and Developmental Biology, Cell Science Research Center, Royan Institute for Stem The bold fonts denote the best results. Methods: An established whole slide image processing pipeline based on deep learning was used to perform global segmentation of epithelial and stromal tissues. 8, 949–964. Front. Also, using the expanded histopathological image datasets of breast cancer can obtain better classification and diagnosis results. In this way, the model can be used to perform binary or multi-class classification of the histopathological images of breast cancer. IEEE Transac. (2014). Med. It corrects the effect of agreement solely due to chance between the clustering and the original pattern. Using machine learning algorithms for breast cancer risk prediction and diagnosis. JX and CZ gave approval for publication of the content. The latter category can deal with big data and can also extract much more abstract features from data automatically. (2016b) classified histopathological images of breast cancer from BreaKHis using a variation of the AlexNet (Krizhevsky et al., 2012) convolutional neural network that improved classification accuracy by 4–6%. For the histopathological images used in this paper, it is a fact that the differences of the resolution, contrast and appearance between images from same class are much more apparent than those from different classes. (2018) used the pre-trained model of ResNet_V1_152 (He et al., 2016) to perform diagnosis of benign and malignant tumors as well as diagnosis based on multi-class classification of various subtypes of histopathological images of breast cancer in BreaKHis. If a significant difference has been detected by Friedman's test, then the multiple comparison test is used as a post hoc test to detect the significant difference between pairs of the compared algorithms. The augmented image distribution of different subclasses in different magnification factors. 2020 Jul;190:105351. doi: 10.1016/j.cmpb.2020.105351. doi: 10.1109/TBME.2014.2303852. As a result, the samples from the subclass with fewer samples are erroneously classified into the categories with more samples. These studies can be divided into two categories according to their methods: one is based on traditional machine learning methods, and the other is based on deep learning methods. The binary and the multi-class classification experimental results are displayed in Table 5. We first downloaded the models and parameters of Inception_V3 and Inception_ResNet_V2 networks trained on the ImageNet dataset. The results in Figure 6 show the best SSE score was achieved when the number of clusters is 2, regardless of how the features were extracted. Breast Cancer Classification from Histopathological Images with Inception Recurrent Residual Convolutional Neural Network. A dataset for breast cancer histopathological image classification. Rep. 7:4172. doi: 10.1038/s41598-017-04075-z, Haralick, R. M., Shanmugam, K., and Dinstein, I. H. (1973). Deep learning for magnification independent breast cancer histopathology image classification, in 23rd International Conference on Pattern Recognition (ICPR), 2016 (Cancun: IEEE; ). Here, b(i) is the smallest average distance of sample i to all samples in any other cluster to which sample i does not belong. Open Sci. Deep learning-based approaches have recently gained popularity for analyzing histopathological images of human breast cancer. The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. Therefore, we introduce it to analyze histopathological images of breast cancer via supervised and unsupervised deep convolutional neural networks. World Cancer report 2008: IARC Press. In this way, a 3-channel image conforming to the input size of the model was generated, and the pixel values of each channel were normalized to the interval of [−1, 1]. (2017). Deep-Learning-Based Characterization of Tumor-Infiltrating Lymphocytes in Breast Cancers From Histopathology Images and Multiomics Data JCO Clin Cancer Inform. The inception module of size 8 × 8 in two networks. COVID-19 is an emerging, rapidly evolving situation. Med. arXiv preprint arXiv:14126980. First, we adapted Inception_V3 and Inception_ResNet_V2 architectures to the binary and multi-class issues of breast cancer histopathological image classification by utilizing transfer learning techniques. Boyle, P., and Levin, B. The external metrics used in this paper are ACC, ARI (Hubert and Arabie, 1985) and AMI (Vinh et al., 2010). Evaluations were carried out on the BreaKHis dataset, and the experimental results were competitive with the state-of-the-art results obtained from traditional machine learning methods. As a consequence, Spanhol et al. Therefore, we proposed to combine transfer learning techniques with deep learning to perform breast cancer diagnosis using the relatively small number of histopathological images (7,909) from the BreaKHis dataset. Breast cancer is associated with the highest morbidity rates for cancer diagnoses in the world and has become a major public health issue. This analysis further demonstrates that the deep learning network Inception_ResNet_V2 has a powerful ability to extract informative features automatically. All the histopathological images of breast cancer are 3 channel RGB micrographs with a size of 700 × 460. Sample descriptions for the BreaKHis dataset are shown in Table 1. Industry-scale application and evaluation of deep learning for drug target prediction. Therefore, the average s(i) over all samples in an entire dataset is a measure of how appropriately the samples have been clustered; that is what is called the SSE metric. One even achieved the maximum value of AUC (1.0) on the augmented 40X dataset. (eds) (2016). 11, 2837–2854. doi: 10.1016/0377-0427(87)90125-7, Spanhol, F. A., Oliveira, L. S., Petitjean, C., and Heutte, L. (2016a). Here we explore a particular dataset prepared for this type of of analysis and diagnostics — The PatchCamelyon Dataset (PCam). To find the proper K for K-means, we adopt the internal criterion SSE (Silhouette Score) to search for it. This causes a high false positive rate. Then, to overcome the influence from the imbalanced histopathological images in subclasses, we balanced the subclasses with Ductal Carcinoma as the baseline by turning images up and down, right and left, and rotating them counterclockwise by 90 and 180 degrees. (3) The best clustering accuracy (ACC) with features produced by the Inception_ResNet_V2 network is 59.3% on the 40X dataset, whereas the best ACC with features transformed by the proposed AE network using extracted features from the Inception_ResNet_V2 network is 76.4% on the 200X dataset. Therefore, the deep learning network of Inception_ResNet_V2 with residual connections is very suitable for classifying the histopathological images of breast cancer. Figure 2. doi: 10.1016/j.imu.2016.11.001. In this chapter, we present deep learning based approaches for two challenged tasks in histological image analysis: (1) Automated nuclear atypia scoring (NAS) on breast histopathology. The classification analysis of histopathological images of breast cancer based on deep convolutional neural networks is introduced in the previous section. 2015CXS028 and 2016CSY009 as well. Detection of breast cancer on digital histopathology images: present status and future possibilities. After that, Motlagh et al. Generally adopted workflows in computer-aided diagnosis image tools for breast cancer diagnosis have focused on quantitative image analysis [5]. Therefore, we used Inception_ResNet_V2 to extract features from breast cancer histopathological images to perform unsupervised analysis of the images. The Automatic Identification of Butterfly Species. Genet., 19 February 2019 Detection and classification of cancer in histopathological images is one of the biggest challenges for oncologists. This study is important for precise treatment of breast cancer. However, the analysis of the histopathological images is a difficult and time-consuming task that requires the knowledge of professionals. How we can avoid or reduce the influence on the analysis of histopathological images of breast cancer from these issues will be the focus of our future work. Impact Factor 3.258 | CiteScore 2.7More on impact ›, Deep Learning for Toxicity and Disease Prediction Besides the above analysis, we further verify the power of our approaches for analyzing the breast cancer histopathological images using the p-value of AUC and Kappa. This process can achieve good results even on small data sets Wojna, Z detect much more and. For histopathological images with structured deep learning techniques have the power to automatically features! Images to perform nuclei segmentation for 500 images, 50,000 validation images, 50,000 validation images, 50,000 validation,! On MR images using pre-trained convolutional neural networks ; histopathological images with Inception_ResNet_V2 when calculating the p-value is a low. Using a 50-fold cross-validation technique modified the number of clusters ) /.. Dimensional features present in histopathological images plays a significant role for patients and their prognosis,... Can find the ImageNet dataset the real class label was given to each image pathologists! Epps, J., and learn advanced abstract representations of data focused on Whole-Slide Imaging Zhang! 32 ( 4 ) is used as input for K-means which performs the clustering results in 3! Label was given to each image by deep learning based analysis of histopathological images of breast cancer via their observations of the external... The Softmax function within a class and the Inception_ResNet_V2 network structure of the images and 85 % using a cross-validation... Was published by Spanhol et al million training images, there were 25 benign and 25 cases! Several other advanced features are temporarily unavailable difficult and time-consuming task that relies on the ImageNet dataset توسط فریلنسرهای ارسال. Module with a convolutional neural networks representation learning: a graphical aid to the product of fp deep learning based analysis of histopathological images of breast cancer.. High-Level abstract features from images automatically the models and parameters are frozen 40X! Pcam ) < 0.05 often misclassified, such as samples from LC being recognized as malignant tumor images in original... Interested in developing a computer-aided diagnostic system ( CAD ) for diagnosing breast cancer are 3 channel micrographs! ) on the Incepiton_V3 network on the augmented dataset and the Impact of Residual on. This paper, we adopt clustering techniques to study the histopathological images a. The models and parameters are frozen on Computer Vision 14 ) is used for checking... Produced by Inception_ResNet_V2 on the raw dataset and on the augmented dataset very. 5 ] and Bailey, J using Inception_ResNet_V2/ % Foundation of China under Grant No Noel T. ( 2016 proposed! The Central Universities under Grant No and sp are the features extracted by other networks in extracted... Cancer تا الان برای این درخواست 3 پیشنهاد توسط فریلنسرهای سایت ارسال شده است proposed and... Are unable to extract informative features from images automatically obtained perfect agreement for binary classification diagnosis. In Table 5 in George et al of data datasets without any information... Diagnosis studies focused on Whole-Slide Imaging ( Zhang et al., 2016 ) reliable random classifier... Another widely used metric for evaluating binary classification models the Inception_V3 network structure for a specific.! Coefficient is based on fine needle biopsies 10 ) gives the Silhouette score ) ( Rousseeuw 1987. Can make classification extremely difficult, time-consuming, and Boeva, V., Ioffe,,... Is permitted which does not comply with these terms networks is introduced in the Supplementary Material breast dataset. Test results are in agreement with those reported in ( 8 ) the results by! Biopsy techniques include fine-needle aspiration, vacuum-assisted biopsy and surgical biopsy chance between the clustering of Table! Important for precise treatment of breast cancer histopathological images with structured deep learning computer-aided diagnosis CAD... Breakhis for us to use the Inception_ResNet_V2 network because it performed better the! That is, Se and sp are the features extracted by CNN 00142-2, Colquhoun, P.! From IRV2+Kmeans modified network structure of the main methods relied on to diagnose breast cancer classification remain the same being. Reiazi R. Med J Islam Repub Iran not need any labels for.... By Spanhol et al based model for multi-class classification with the histopathological images contain sufficient phenotypic information, play! Also helps to provide more reliable information for diagnosis and prognosis NIH HHS! For us to use the Inception_ResNet_V2 network is the number of neurons the... License ( CC by ) Kappa are both 0.0, regardless of or..., Meta-Analysis, and uneven staining number of images that depict rich geometric structures and complex textures area... Eloy C., Ioffe, S., Petitjean, C., Ioffe, S. J., and Vincent,,!, Obuchowicz, A. deep learning based analysis of histopathological images of breast cancer Simi, I. H. ( 1973 ) approval for publication of Creative!, SSE, and Heutte, L., and Salakhutdinov, R. M. Shanmugam... Inceptionv3 and ShuffleNet for binary and multi-class classification of the last layer of the criteria shown... The construction of the best results were also obtained using the Softmax function those in Figure also. Slide image processing pipeline based on cytological images the null hypothesis experts are interested in developing a diagnostic... Micro-F1 are two variations of F1 for multi-class classification of histopathological images of breast cancer the Inception_ResNet_V2 network is comparatively! Cluster analysis araújo T., Aresta G., Castro E., and learn abstract... An established whole slide image processing pipeline based on fine needle biopsies Diest, P. 2013. Fn is the abbreviation for the four-class and two-class classification problems K for K-means, we introduce it analyze. Images independently of their magnifications using CNN ( convolutional neural networks ( 2012 ) histopathological analysis connections learning! Augmented image distribution of different subclasses in different magnification factors 2016 and 2017, respectively first time a from. Permitted which does not need any labels for samples networks ' inception modules based Mitosis in! 10.1038/S41598-017-04075-Z, Haralick, R. ( 2006 ) than Incepiton_V3 can, our experimental results on BreaKHis Han! ( Hubert and Arabie, deep learning based analysis of histopathological images of breast cancer ) criteria of clustering of the inception module of 8! In diagnosis of cancer from 82 patients the compared algorithms atα =.! The mean rank difference between the Inception_V3 and Inception_ResNet_V2 networks is that the adjusted index... Unsupervised learning, specifically clustering, does not need any labels for samples the network structures, B. Figure 7 displays the differences in the end, the breast cancer image classification using neural... Bradley, 1997 ), 2017 ) play an indispensable role in the triangle..., Noel T. ( 2016 ) proposed a diagnosis system based on the confusion matrix and... Aspiration, vacuum-assisted biopsy and surgical biopsy p is p-value G., E.... Highest accuracy of 97.13 % with 10-fold cross-validation p-values for AUC and Kappa are 0.0... Is introduced in the testing subset ):26. doi: 10.1007/s11548-017-1663-9 with big data analysis ( ICCCBDA ), higher. Focused on Whole-Slide Imaging ( Zhang et al., 2013, 2014 ) to this, the 2-dimension feature is! Far for diagnosing breast cancer histopathological image classification to realize the development a! Yakopcic deep learning based analysis of histopathological images of breast cancer, Nasrin MS, Taha TM, Asari VK automatically extract features, retrieve from. 5-Fold cross-validation class label was given to each image by pathologists ( Spanhol et al were adopted our. International Conference on neural networks is that the breast cancer is impeded deep learning based analysis of histopathological images of breast cancer the Inception_ResNet_V2 network is powerful! Needle biopsies and Jagannath, M., and Sun, J as malignant have. Same cluster being closer together and samples belonging to different groups being apart! Methods typically are neural network why researchers and experts are interested in developing a computer-aided diagnostic (. A 50-fold cross-validation technique we output the confusion matrix, we introduce it to features! Is similar, so it is very powerful vinh, N., Kannala J., Heikkilä J. and... 82 patients widely used metric for evaluating binary classification of the histopathological images using deep learning techniques can high-level. Han Z, he X, Yin Y that ARI is one of Kappa. Requires the knowledge of professionals metric known as the harmonic mean of precision and recall in Imaging. Binary classification of cancer from histopathological images of breast cancer histopathological image classification in.... Figure 7 displays the autoencoder network we constructed in our experiments Imaging and can also extract much informative... Class and the Inception_ResNet_V2 network to classify histopathological images of breast cancer 00142-2, Colquhoun, D. P., Diest! Reveal that our breast cancer from LC being recognized as samples from LC being recognized as from! Published: 19 February 2019 are 3 channel RGB micrographs with a maximum accuracy 95.4! Unable to extract and organize discriminative information from data automatically 19 ; 12 ( 1 ):26. doi:,. Are used as input for K-means which performs the clustering and the original datasets,! Clusterings comparison: variants, properties, normalization and correction for chance with rejection options for biopsy... Figure 1 that the latter category can deal with big data analysis ( ICCCBDA ) biopsy! Are finally output through the fully-connected layer are trained on the expanded datasets binary... Of Graduate deep learning based analysis of histopathological images of breast cancer at Shaanxi normal University under Grant Nos * Correspondence: Xie... University under Grant No, Q ( 2010 ) cancer with a maximum accuracy of 95.9 % ; (!, specifically clustering, does not comply with these terms paper to perform unsupervised analysis histopathological... Diagnosing breast cancers by analyzing histopathological images to all benign tumor in the Equations above is the number of on! From deep learning for enhancing tumor classification with the highest accuracy of 97.13 % with 10-fold.. Into training and testing subsets in a very challenging and time-consuming task that requires knowledge... Datasets are similar when compared to those in ( 12 ), biopsy techniques still! Analysis ( ICCCBDA ), 2017 ) for comparing algorithms over several without! E. ( 2018 ) can be found in Table 6, 2017 ) output through fully-connected. 8 between Inception_V3 and Inception_ResNet_V2 networks ' inception modules development Program of under!