Convolution Neural Networks for diagnosing colon and lung cancer histopathological images
CConvolution Neural Networks for diagnosing colonand lung cancer histopathological images
Sanidhya Mangal
Engineerbabu [email protected]
Aanchal Chaurasia
Engineerbabu [email protected]
Ayush Khajanchi
Emorphis Technologies [email protected]
Abstract
Lung and Colon cancer are one of the leading causes of mortality and morbidityin adults. Histopathological diagnosis is one of the key components to discerncancer type. The aim of the present research is to propose a computer aideddiagnosis system for diagnosing squamous cell carcinomas and adenocarcinomasof lung as well as adenocarcinomas of colon using convolutional neural networksby evaluating the digital pathology images for these cancers. Hereby, renderingartificial intelligence as useful technology in the near future. A total of 2500 digitalimages were acquired from LC25000 dataset containing 5000 images for eachclass. A shallow neural network architecture was used classify the histopathologicalslides into squamous cell carcinomas, adenocarcinomas and benign for the lung.Similar model was used to classify adenocarcinomas and benign for colon. Thediagnostic accuracy of more than
97 % and
96 % was recorded for lung and colonrespectively.
According to the World health organization(WHO) cancer is the leading cause of morality in theworld. Lung cancer is the most commonly diagnosed cancer ( . of the total cases) and theleading cause of cancer death ( . of the total cancer deaths), while colorectal cancer contributesto ( . ) for mortality Bray et al.. A rising trend has been reported around the globe for theincidence of malignant tumors which may be attributed to an increase in population. Malignancy canoccur to any age group by histopathological type is diagnosed mostly in elderly age group of 50-60years Arslan et al.. It has been predicted that cancer mortality might bump up to
60 % untill 2035Araghi et al..Lung cells become cancerous when they mutate to grow uncontrollably and form a cluster called atumor. Ame The worldwide increase in lung cancer has been attributed to different factors but mainlyto the exposure to breathing dangerous or toxic substances and the rise in the number of aged peoplein the society. The symptoms, however, are not likely to be observed until it has spread to the otherparts of the body, which makes it harder to treat it.Although lung cancer can occur in people who have never smoked, it usually is the greatest risk forpeople who do. Adenocarcinoma and squamous cell carcinoma are the most commonly occurringtypes of lung cancer while the other histological types include small as well as large cell carcinomas.Adenocarcinoma of lung cancer usually occurs in current or former smokers but is also prevalentin non-smokers. This is more prone to occur in women and youth and is found in the outer parts of
Preprint. Under review. a r X i v : . [ ee ss . I V ] S e p he lungs even before it spreads. Squamous cell carcinomas are also associated with the history ofsmoking. Small and Large cell carcinoma, on the other hand, can develop in any part of the lung andhas the tendency to grow and unroll rapidly making it harder to medicate.Colon, the final part of our digestive system, when host cancerous cells, could cause colon cancer.Colon cancer is not age-specific but it typically affects older adults. It usually begins as small,noncancerous (benign) clumps of cells called polyps that form on the inner side of the colon. Overtime, some of these polyps can develop to cause colon cancers.In most colon cancers, a tumor is formed when healthy cells in the lining of the colon or rectum growuncontrollably. Adenocarcinomas of the colon or rectum develop in the lining of the large intestineoccuring in the epithelial cells and then spread to the other layers. Mucinous adenocarcinomas andsignet ring cell adenocarcinoma are two less common subtypes of adenocarcinoma but are aggressiveand difficult to medicate. The changes can occur over the years in one’s body which is reliant onfactors such as gender, ethnicity, age, smoking patterns, and socio-economic conditions. However, ifa person has some unusual inherited syndrome, the changes can transpire in a small duration of somemonths.The aim of the research paper revolves around measuring the potency of the proposed algorithm ofthe Convolutional Neural Network (CNN) LeCun et al. to detect the common types of lung and coloncancer in the human body. The architecture of the algorithm keeps in mind the patterns of the neuronsand their connectivity inside the human brain which also attributes for it’s low pre-processing requiredin comparison to other classification algorithms. The ability of the algorithm to learn characteristicsoverpowers the primitive method wherein filters are hand-engineered. The ConvNet takes input inthe form of images attributes weight (learnable weights and biases) to multiple features in the image,and is able to differentiate one image from the other. We use Histopathology slides Gurcan et al. asa dataset, since the preparation process preserves the underlying tissue architecture it provides aneclectic view of disease and its effect on tissues. Additionally,histopathology image renders as a‘gold standard’ in diagnosing almost distinct cancer types Rubin et al..In the later section the paper is organized in the following manner: Section 2 provides an insightabout previously explored research in the current domain. Section 3 provided a brief introductionabout LC25000 dataset. Section 4 covers a short introduction on CNN, In addition to this Section 5elaborates the cnn architecture used in training both the models. Section 6 reports all the experimentalresults and findings. Finally, Section 7 concludes our experiment while presenting some insights onfuture work. Doi explored the potential for automatic image processing around 4 decades ago but it is stillchallenging due to complexity of images to analyze. Back then feature extraction was a key step inadopting machine learning based computer-aided diagnosis (CAD). Different ontologies of cancerhas been investigated in, Brake te et al., Beller et al., Yin et al., Aerts et al., Eltonsy et al., Weiet al., Hawkins et al., Barata et al., Barata et al., Han et al., Sadeghi et al., Zikic et al., Meieret al.. Moreover, Munir et al. provides a detailed overview on cancer diagnosis by conductingexperiments on several deep learning techniques. It also provides a comparison of various predominantarchitectures for each technique.Also, Coudray et al. trained an inceptionv3 Szegedy et al. model for classification and mutationfrom non–small cell lung images of adenocarcinomas and squamous cell carcinomas achieving amean area under curve(auc) of 0.97. They also mutated the result for ten most common genes forlung adenocarcinomas. Similarly Ardila et al. predicts the risk of lung cancer using deep learningtechniques by computing prior and current tomography of the patients. Lakshmanaprabu et al.explored the CT scans using deep neural networks and linear discriminant analysis for automateddiagnosis. Sirinukunwattana et al. used spatially constrained CNN to perform nuclei detection andclassification of cancerous tissues in colon histology images.Recently, Abbas et al. presented a comparative study on histopathology diagnosis on squamous cellcarcinomas using CNNs. It compares various CNN architectures such as AlexNet, VGG-16, ResNetachieving an F-1 score of 0.97. Similarly, Bukhari et al. presents a comparative analysis on colonicadenocarcinomas using variations of ResNet architecture achieving a baseline accuracy of
93 % .2 LC25000 Dataset
A brief introduction on dataset is provided followed by all the data preprocessing steps. TheLC25000 Dataset Borkowski et al. contains microsopic images of lung and colon. The datasetcan be bifuragted into five different classes namely, lung adenocarcinomas, lung squamous cellcarcinomas, lung benign, colon adenocarcinomas and colon benign each containing images.Figure 1 describes some sample images corresponding to above mentioned classes. Original datasetcontains only images lung and images of colon with pixel size of , later it wasconverted into square of pixels. Augmentor was used to expand the dataset into images with the help of rotation and flips. (a) lung adenocarcinomas (b) lung squamous cell carcinomas (c) lung benign(d) colon adenocarcinomas (e) colon benign
Figure 1: (a) and (b) is an example image of adenocarcinomas and squamous cell carcinomas cancertypes for the lung, (c) represents the benign histopathology of lung. Similarly in (d) adenocarcinomascancer for colon is described where in (e) illustrates benign class for colonBefore feeding augmented data, it undergo some preprocessing. Initially data was sampled into and datapoints for training and test set respectively for each class using random samplingproposed in Yates. Furthermore images were resized to pixels, along with some randomizedshear, zoom transformation followd by normaliztion of images.
Image classification is a challenging task for the visual content, particularly microscopic imagesfor example histopathological images due to high convolution of inter-intraclass dependencies. Theunderlying structures are complex and interwoven due to similar structural morphological textures.Figure 1 presents some of the complex textures present in histopathology of images. Deep learning isprevalent due to its ability to learn features directly from the input, providing us a window to avoidarduous feature extraction processes Bengio et al.. One of the key features of deep learning is todiscover abstract level features and then deep dive for extracting structural semantics in the featuremap. In recent years, deep learning, especially CNNs has proven to be an effective tool for classifyingand diagnosing medical images Shen et al., Kermany et al., Lee et al., Suzuki In nutshell, CNNcontains multiple trainable layers which could be stacked together along with a supervised classifierto learn feature maps from the given input data feed. Input data feed could be either digital or signaldata such as audio, video, images and time-series. For example, upon considering a coloured image itis a feature map of 3D tensor, i.e., a 2D tensor for each colour channel.3NN architectures are composed of mainly three layers which are: convolution layer, max poolinglayer and fully connected layer or dense layer. These layers could be stacked in multiple combinationsto produce a CNN. An example of typical CNN is show in Figure 2.Figure 2: Describes one of the most commonly used cnn architectures, Alom et al.In a typical CNN, convolution layer acts as a key component for any given architecture. Convolutionlayers compute a dot product between weights and input signal connected to that local region. The setof weights which are convoluted along the input vector is called kernel or filters . Each filter issmall but extends across the full depth of the input volume. For image inputs a typical size of filter isgenerally (3x3, 5x5, 8x8). These weights are shared across neurons so that filters can learn all thegeometrical structures from the image. The distance between applications of these filters is termed as stride . If hyperparameter stride is smaller than filter size than overlapping convolutions are appliedto the image.It is a common practice to insert a pooling in between two convolution layers in order to downsamplethe image along volume component, This is crucial to reduce progressively the spatial size of therepresentation. Thus, reducing the number of parameters and computations required by the networkhelps in the overfitting control. The pooling operation resizes the images along height and widthdiscarding activation. In practice, max pooling operation which provides a window of selecting themaximum value from input patch among neighborhoods has shown better results Scherer et al..In fully connected layer or dense layer , full connection is established between activations of inputand their activation. Computation is done with the help of matrix multiplication along with successivebias offset. The last fully connected layer contains the final output such as probability density or logit values Spanhol et al., Krizhevsky et al..
In order to classify the (citeauthor dataset name) dataset we constructed the deep CNN with followinglayers and parameters:
Input Layer
This layer is used to load data and feed it to the first convolution layer, In our case theinput is an image of size 150x150 pixels with colour channels which is 3 for RGB.
Convolution Layer
This layer is used to convolve the input image with trainable filters to learnthe geo spatial structure of images, this model contains three convolution layers with filter size 3x3,stride set to 2 and padding kept the same. First layer contains 32 filters, followed by two layers with64 filters each and they are initialized with Gaussian distribution. In addition to this, ReLU activationis applied for non linear operation to improve the performance Behnke
Pooling Layer
Pooling operation is used for downsampling the output images received from theconvolution layer. There is one pooling layer after each convolution layer with pooling size of 2,padding set to valid. All the pooling layers use the most common max pooling operation.
Flatten Layer
This layer is used to convert the output from the convolution layer into a 1D tensorto connect a dense layer or fully connected layer.4 ully connected layer or dense layer
These layers treat the input as a simple vector and producean output in a form of vector. Two dense layers are used in this model, first one contains 512 neuronsand last one contains 3 and 2 neurons for lung and colon cancer respectively depending on theinput class. Output from the last fully connected layer could be activated with the help of softmaxactivation.
Sof tmax ( x i ) = e x i (cid:80) j e x j (1) Dropout Layer
In order to prevent overfitting of the model layers we use a dropout layer in betweenfully connected layers which randomly drops neurons from both visible and hidden layers Srivastavaet al..Table 1 Illustrates the parameters of the layers, where CONV+POOL stands for convolution layerfollowed by a pooling layer and FC by fully connected layer or dense layer.Table 1: Summary of CNN LayersLayers1 2 3 4 5Type CONV+POOL CONV+POOL CONV+POOL FC FCChannels 32 64 64 512 3 or 2Filter Size 3x3 3x3 3x3 - -Convolution Strides 2x2 2x2 2x2 - -Pooling Size 2x2 2x2 2x2 - -Pooling Strides 1x1 1x1 1x1 - -For all the CNN modes a similar training protocol was used, purely supervised in nature. TheRMSprop method proposed by Tieleman, Hinton with backpropagation was used to computegradient and a mini batch size of 32 was used to update network weights, with starting learning rateof 10-4, in conjunction with ρ = 0 . and (cid:15) = 1 e − . Categorical cross entropy loss 2 is used toensure that performance of the model is maintained throughout the training process. The CNN wastrained for 100 iterations. E entropy = − N (cid:88) n c (cid:88) k t nk ln y nk (2) To ensure that classifiers generalize well, the data was split into three categories, with 80-10-10 of data into training, testing and validation set respectively into distinct sets. This protocolwas applied independently for both lung and colon cancer images. When discussing medicalimage processing it can be evaluated in two ways, first one is at patient level, i.e., determining theamount of images classified correctly for each patient. Secondly, it can be evaluated at the imagewhere we calculate the number of cancer images classified correctly. In LC25000 dataset 3 noinformation about patients was provided hence we decided to move forward with a later methodto evaluate the model performance. All the CNN models were trained on Google’s Colab usingTensorFlow framework Abadi et al.. These models would be made available in h5 format at https://github.com/sanidhyamangal/lung_colon_cancer_research . Training each model tookaround 45 minutes.Deep Learning techniques are one of the advanced machine learning techniques which do not requirethe design of feature extraction by domain experts but model learns by itself. We can learn the featuredetectors learned by models, considering the weights learned by feature maps. Figure 3 describesfeature maps learned by all the convolution layers for both the models. We can visualize the filters atimage level and also filters resemble like Gabor filters( Fogel, Sagi, Bovik et al., Zeiler, Fergus).To better assess performance metrics, Figure 4 visualizes the plot between epochs vs loss and accuracy.It is quite transparent that there are jitters in all the accuracy and loss graphs due to the dropout layerwhich helps the neural network generalize. However, there is a slight aberration in the plots of colon,5 a) colon filter 1 (b) colon filter 2 (c) colon filter 3(d) lung filter 1 (e) lung filter 2 (f) lung filter 3
Figure 3: Feature maps learned by convolution network layers, (a), (d) describes filters form firstconvolution layer for colon and lung models respectively, Similarly, (b), (e), (c), (f) illustrates thefilters from second and third convolution layers respectively. (a) Colon Loss (b) Colon Accuracy(c) Lung Loss (d) Lung Accuracy
Figure 4: Delineates the accuracy and loss metrics plot over steps. (a), (b) reports the loss andaccuracy metrics over all the 100 steps for the colon model. Similarly, loss and accuracy for lungmodel is described in (c), (d)i.e., validation loss increases initially upto 5 epochs then starts to converge. Also, it can be inferredthat there are large spikes in both accuracy and loss for the colon model at around th and th epoch. For both the models it could also be extrapolated that both the models took around 20 epochsto converge. 6able 2 reports the performace metrics for both the models, in both the modes, i.e., training andvalidation at image level. Table 2: Performance metrics of all the modelsType Rule Accuracy LossLung Training . .
832 30(6149) %
Validation . .
114 50(5719) %
Colon Training . . Validation . . In this paper, we have presented a set of experiments conducted on the LC25000 dataset using adeep learning approach. We have shown that we could use a shallow CNN architecture, that hasbeen designed for classifying color images of objects, and adapt it to the classification of lungand colon histopathological images. We have also proposed a training and evaluation strategy fortraining the CNN architecture, it allows to deal with the high-resolution of these textured imageswithout converting those images to low-resolution images. Our experimental results obtained onthe LC25000 showed improved accuracy obtained by CNN when compared to traditional machinelearning models and deep convolutional neural networks models leveraging transfer learning trainedon the same dataset but with state of the art texture descriptors. Future work can explore differentCNN architectures and the optimization of the hyperparameters. Also, strategies to apply neural styletransfer for generating interclass images for different histopathology. In addition to this generativemodels could be used to generate histopathological images for visualizing and exploring mutationson different ontologies.
Acknowledgments and Disclosure of Funding
We would like to acknowledge Mayank Pratap Singh, Aditi Chaurasia, Sumit Yadav, and PrachiBundela for helpful discussions. Ravinder Singh shared his script for train test split with us andmuch-needed support with L A TEX typesetting. We would like to thank the developers of TensorFlow.We would also like to thank Engineerbabu IT Services PVT LTD. for providing computationalresources.
References
About Lung Cancer: Lung Cancer Overview. ????
Abadi Martín, Agarwal Ashish, Barham Paul, Brevdo Eugene, Chen Zhifeng, Citro Craig, Corrado Greg S.,Davis Andy, Dean Jeffrey, Devin Matthieu, Ghemawat Sanjay, Goodfellow Ian, Harp Andrew, Irving Geoffrey,Isard Michael, Jia Yangqing, Jozefowicz Rafal, Kaiser Lukasz, Kudlur Manjunath, Levenberg Josh, ManéDandelion, Monga Rajat, Moore Sherry, Murray Derek, Olah Chris, Schuster Mike, Shlens Jonathon, SteinerBenoit, Sutskever Ilya, Talwar Kunal, Tucker Paul, Vanhoucke Vincent, Vasudevan Vijay, Viégas Fernanda,Vinyals Oriol, Warden Pete, Wattenberg Martin, Wicke Martin, Yu Yuan, Zheng Xiaoqiang . TensorFlow:Large-Scale Machine Learning on Heterogeneous Systems. 2015. Software available from tensorflow.org.
Abbas Mohammad Ali, Bukhari Syed Usama Khalid, Syed Asmara, Shah Syed Sajid Hussain . The Histopatho-logical Diagnosis of Adenocarcinoma & Squamous Cells Carcinoma of Lungs by Artificial intelligence: Acomparative study of convolutional neural networks // medRxiv. 2020.
Aerts Hugo JWL, Velazquez Emmanuel Rios, Leijenaar Ralph TH, Parmar Chintan, Grossmann Patrick, CarvalhoSara, Bussink Johan, Monshouwer René, Haibe-Kains Benjamin, Rietveld Derek, others . Decoding tumourphenotype by noninvasive imaging using a quantitative radiomics approach // Nature communications. 2014.5, 1. 1–9.
Alom Md. Zahangir, Taha Tarek, Yakopcic Chris, Westberg Stefan, Sidike Paheding, Nasrin Mst, Hasan Mah-mudul, Essen Brian, Awwal Abdul, Asari Vijayan . A State-of-the-Art Survey on Deep Learning Theory andArchitectures // Electronics. 03 2019. 8. 292. raghi Marzieh, Soerjomataram Isabelle, Jenkins Mark, Brierley James, Morris Eva, Bray Freddie, ArnoldMelina . Global trends in colorectal cancer mortality: projections to the year 2035 // International journal ofcancer. 2019. 144, 12. 2992–3000. Ardila Diego, Kiraly Atilla P, Bharadwaj Sujeeth, Choi Bokyung, Reicher Joshua J, Peng Lily, Tse Daniel,Etemadi Mozziyar, Ye Wenxing, Corrado Greg, others . End-to-end lung cancer screening with three-dimensional deep learning on low-dose chest computed tomography // Nature medicine. 2019. 25, 6.954–961.
Arslan Necmi, Yilmaz Ahmet, Firat Ugur, Tanriverdi Mehmet Halis . Analysis of cancer cases from DicleUniversity Hospital; ten years’ experience Analysis of cancer cases // JOURNAL OF CLINICAL ANDANALYTICAL MEDICINE. 2018. 9, 2. 102–106.
Barata Catarina, Marques Jorge S, Rozeira Jorge . A system for the detection of pigment network in dermoscopyimages using directional filters // IEEE transactions on biomedical engineering. 2012. 59, 10. 2744–2754.
Barata Catarina, Ruela Margarida, Mendonça Teresa, Marques Jorge S . A bag-of-features approach for theclassification of melanomas in dermoscopy images: The role of color and texture descriptors // Computervision techniques for the diagnosis of skin cancer. 2014. 49–69.
Behnke Sven . Hierarchical neural networks for image interpretation. 2766. 2003.
Beller Michael, Stotzka Rainer, Müller Tim Oliver, Gemmeke Hartmut . An example-based system to support thesegmentation of stellate lesions // Bildverarbeitung für die Medizin 2005. 2005. 475–479.
Bengio Yoshua, Courville Aaron, Vincent Pascal . Representation learning: A review and new perspectives //IEEE transactions on pattern analysis and machine intelligence. 2013. 35, 8. 1798–1828.
Borkowski Andrew A., Bui Marilyn M., Thomas L. Brannon, Wilson Catherine P., DeLand Lauren A., MastoridesStephen M.
Lung and Colon Cancer Histopathological Image Dataset (LC25000). 2019.
Bovik Alan C., Clark Marianna, Geisler Wilson S.
Multichannel texture analysis using localized spatial filters //IEEE transactions on pattern analysis and machine intelligence. 1990. 12, 1. 55–73.
Brake Guido M te, Karssemeijer Nico, Hendriks Jan HCL . An automatic method to discriminate malignantmasses from normal tissue in digital mammograms1 // Physics in Medicine & Biology. 2000. 45, 10. 2843.
Bray Freddie, Ferlay Jacques, Soerjomataram Isabelle, Siegel Rebecca L, Torre Lindsey A, Jemal Ahmedin .Global cancer statistics 2018: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in185 countries // CA: a cancer journal for clinicians. 2018. 68, 6. 394–424.
Bukhari Syed Usama Khalid, Asmara Syed, Bokhari Syed Khuzaima Arsalan, Hussain Syed Shahzad, ArmaghanSyed Umar, SHAH SYED SAJID HUSSAIN . The Histological Diagnosis of Colonic Adenocarcinoma byApplying Partial Self Supervised Learning // medRxiv. 2020.
Coudray Nicolas, Ocampo Paolo Santiago, Sakellaropoulos Theodore, Narula Navneet, Snuderl Matija, FenyöDavid, Moreira Andre L, Razavian Narges, Tsirigos Aristotelis . Classification and mutation prediction fromnon–small cell lung cancer histopathology images using deep learning // Nature medicine. 2018. 24, 10.1559–1567.
Doi Kunio . Computer-aided diagnosis in medical imaging: historical review, current status and future potential// Computerized medical imaging and graphics. 2007. 31, 4-5. 198–211.
Eltonsy Nevine H, Tourassi Georgia D, Elmaghraby Adel S . A concentric morphology model for the detectionof masses in mammography // IEEE transactions on medical imaging. 2007. 26, 6. 880–889.
Fogel Itzhak, Sagi Dov . Gabor filters as texture discriminator // Biological cybernetics. 1989. 61, 2. 103–113.
Gurcan Metin N, Boucheron Laura E, Can Ali, Madabhushi Anant, Rajpoot Nasir M, Yener Bulent . Histopatho-logical image analysis: A review // IEEE reviews in biomedical engineering. 2009. 2. 147–171.
Han Fangfang, Wang Huafeng, Zhang Guopeng, Han Hao, Song Bowen, Li Lihong, Moore William, Lu Hongbing,Zhao Hong, Liang Zhengrong . Texture feature analysis for computer-aided diagnosis on pulmonary nodules// Journal of digital imaging. 2015. 28, 1. 99–115.
Hawkins Samuel H, Korecki John N, Balagurunathan Yoganand, Gu Yuhua, Kumar Virendra, Basu Satrajit, HallLawrence O, Goldgof Dmitry B, Gatenby Robert A, Gillies Robert J . Predicting outcomes of nonsmall celllung cancer using CT image features // IEEE access. 2014. 2. 1418–1426. ermany Daniel S, Goldbaum Michael, Cai Wenjia, Valentim Carolina CS, Liang Huiying, Baxter Sally L,McKeown Alex, Yang Ge, Wu Xiaokang, Yan Fangbing, others . Identifying medical diagnoses and treatablediseases by image-based deep learning // Cell. 2018. 172, 5. 1122–1131. Krizhevsky Alex, Sutskever Ilya, Hinton Geoffrey E . Imagenet classification with deep convolutional neuralnetworks // Advances in neural information processing systems. 2012. 1097–1105.
Lakshmanaprabu SK, Mohanty Sachi Nandan, Shankar K, Arunkumar N, Ramirez Gustavo . Optimal deeplearning model for classification of lung cancer on CT images // Future Generation Computer Systems. 2019.92. 374–382.
LeCun Yann, Bengio Yoshua, others . Convolutional networks for images, speech, and time series // Thehandbook of brain theory and neural networks. 1995. 3361, 10. 1995.
Lee June-Goo, Jun Sanghoon, Cho Young-Won, Lee Hyunna, Kim Guk Bae, Seo Joon Beom, Kim Namkug . Deeplearning in medical imaging: general overview // Korean journal of radiology. 2017. 18, 4. 570–584.
Meier Raphael, Bauer Stefan, Slotboom Johannes, Wiest Roland, Reyes Mauricio . A hybrid model for multimodalbrain tumor segmentation // Multimodal Brain Tumor Segmentation. 2013. 31. 31–37.
Munir Khushboo, Elahi Hassan, Ayub Afsheen, Frezza Fabrizio, Rizzi Antonello . Cancer diagnosis using deeplearning: a bibliographic review // Cancers. 2019. 11, 9. 1235.
Rubin Raphael, Strayer David S, Rubin Emanuel, others . Rubin’s pathology: clinicopathologic foundations ofmedicine. 2008.
Sadeghi Maryam, Lee Tim K, McLean David, Lui Harvey, Atkins M Stella . Detection and analysis of irregularstreaks in dermoscopic images of skin lesions // IEEE transactions on medical imaging. 2013. 32, 5. 849–861.
Scherer Dominik, Müller Andreas, Behnke Sven . Evaluation of pooling operations in convolutional architecturesfor object recognition // International conference on artificial neural networks. 2010. 92–101.
Shen Dinggang, Wu Guorong, Suk Heung-Il . Deep learning in medical image analysis // Annual review ofbiomedical engineering. 2017. 19. 221–248.
Sirinukunwattana Korsuk, Raza Shan E Ahmed, Tsang Yee-Wah, Snead David RJ, Cree Ian A, Rajpoot Nasir M .Locality sensitive deep learning for detection and classification of nuclei in routine colon cancer histologyimages // IEEE transactions on medical imaging. 2016. 35, 5. 1196–1206.
Spanhol Fabio Alexandre, Oliveira Luiz S, Petitjean Caroline, Heutte Laurent . Breast cancer histopathologicalimage classification using convolutional neural networks // 2016 international joint conference on neuralnetworks (IJCNN). 2016. 2560–2567.
Srivastava Nitish, Hinton Geoffrey, Krizhevsky Alex, Sutskever Ilya, Salakhutdinov Ruslan . Dropout: a simpleway to prevent neural networks from overfitting // The journal of machine learning research. 2014. 15, 1.1929–1958.
Suzuki Kenji . Overview of deep learning in medical imaging // Radiological physics and technology. 2017. 10, 3.257–273.
Szegedy Christian, Vanhoucke Vincent, Ioffe Sergey, Shlens Jon, Wojna Zbigniew . Rethinking the inceptionarchitecture for computer vision // Proceedings of the IEEE conference on computer vision and patternrecognition. 2016. 2818–2826.
Tieleman T., Hinton G.
Lecture 6.5—RmsProp: Divide the gradient by a running average of its recent magnitude.2012.
Wei Jun, Sahiner Berkman, Hadjiiski Lubomir M, Chan Heang-Ping, Petrick Nicholas, Helvie Mark A, RoubidouxMarilyn A, Ge Jun, Zhou Chuan . Computer-aided detection of breast masses on full field digital mammograms// Medical physics. 2005. 32, 9. 2827–2838.
Yates Daniel . The practice of statistics : TI-83/84/89 graphing calculator enhanced. New York: W.H. Freeman,2008.
Yin Fang-Fang, Giger Maryellen L, Doi Kunio, Vyborny Carl J, Schmidt Robert A . Computerized detection ofmasses in digital mammograms: Automated alignment of breast images and its effect on bilateral-subtractiontechnique // Medical Physics. 1994. 21, 3. 445–452.
Zeiler Matthew D., Fergus Rob . Visualizing and Understanding Convolutional Networks // Lecture Notes inComputer Science. 2014. 818–833. ikic Darko, Glocker Ben, Konukoglu Ender, Criminisi Antonio, Demiralp Cagatay, Shotton Jamie, ThomasOwen M, Das Tilak, Jena Raj, Price Stephen J . Decision forests for tissue-specific segmentation of high-gradegliomas in multi-channel MR // International Conference on Medical Image Computing and Computer-Assisted Intervention. 2012. 369–376.. Decision forests for tissue-specific segmentation of high-gradegliomas in multi-channel MR // International Conference on Medical Image Computing and Computer-Assisted Intervention. 2012. 369–376.