Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. PubMed Google Scholar. In [24], a three-stage compression pipeline is proposed: prune the important connections of the network, then achieve weight sharing by quantizing the weights, and finally apply Huffman coding to further remove the redundancy. 13 shows the recognition accuracies by using our channel pruning and DNS together. In our work, we use the activation factors si (i=1,2,...,C) obtained by SE block as channel weights in assisting the model compression. In: Systems, Man, Q7 and Cybernetics (SMC), 2017 IEEE International Conference On. 2014; 61(5):1400–11. and breast cancer sub-types including Tissue Micro Array (TMA) database ([23]) and BreaKHis (The Breast Cancer Histopathological Images) ([48]). In this paper, we implemented deep neural networks ResNet18, InceptionV3 and ShuffleNet for binary classification of breast cancer in histopathological images. Reinhard E, Adhikhmin M, Gooch B, Shirley P. Color transfer between images. Diagnosis of the type of breast cancer using histopathological slides and Deep CNN features. A Dataset for Breast Cancer Histopathological Image Classification. Exemplar images collected from (a) BreaKHis dataset and (b) BACH dataset. Classification of breast cancer histology images using convolutional neural networks. histopathological breast image classification using Tamura features. NIH In this section, we propose our breast cancer histopathology image classification scheme. Our proposed scheme in this work can be used in breast cancer auxiliary diagnostic scenario, and realize workload reducing and diagnosis quality promoting talked above. In our work, the Inception module [28], residual network [18], and Batch Normalization (BN) techniques [29] are combined together to ensure recognition performance. Article  Spanhol FA, Oliveira LS, Petitjean C, Heutte L: Breast cancer histopathological image classification using convolutional neural networks. Generally, great efforts and effective expert domain knowledge are required to design appropriate features for this type of method. Then the feature maps X are reweighted to $$\tilde {\textbf {X}}$$ : where $$\tilde {\textbf {X}} = \left [\tilde {\textbf {x}}_{1},\tilde {\textbf {x}}_{2},...,\tilde {\textbf {x}}_{C}\right ]$$, and X=[x1,x2,...,xC]. In the structure, 1 ×1 convolutions are used to compute reductions before the expensive higher dimensional filters: 3 × 3 and 5 ×5 convolutions. Besides, the BN technique is adopted to allow the utilization of much higher learning rates and be less careful about initialization by normalizing layer inputs, which ensures a high robustness of our model. 10, a channel pruning example with different R (1 to 4) under the same target pruning ratio O=80% is shown to further analyze the relationship between accuracy and R. With the increasing of R, the model accuracy is improved accordingly and the pruning proportion X for each loop drops. The optimized compact hybrid model achieves comparable results when compared with Table 3 and Table 4. Deep-Learning-Based Semantic Labeling for 2D Mammography and Comparison of Complexity for Machine Learning Tasks. CNN classifies the histopathological images of breast cancer with independent magnification, thus obtaining a higher recognition rate[10, 24]. First, a hybrid CNN architecture is designed, which contains a global model branch and a local model branch. In the diagnosis of preinvasive breast cancer, some of the intraductal proliferations pose a special challenge. Breast cancer remains the most common type of cancer and the leading cause of cancer-induced mortality among women with 2.4 million new cases diagnosed and 523, 000 deaths per year. The excitation operation can explicitly model interdependencies between channels. We propose another different channel pruning method, which can accurately control how many channels are pruned. 7. The 200 × magnification factor shows the best results among performances obtained with different magnification levels under 0.4 False Positive Rate (FPR). BMC Medical Informatics and Decision Making 2020 Nov;4:1039-1050. doi: 10.1200/CCI.20.00110. This is why researchers and experts are interested in developing a computer-aided diagnostic system (CAD) for diagnosing histopathological images of breast cancer… (a) Adopted inception architecture. In work [11], the authors provide two strategies to generate the training samples: sliding window allowing 50% of overlap between patches; random extraction strategy with a fixed arbitrary number of patches (such as 1000) from each input image. Each pixel of these regions (the remaining tissue is considered normal) has a corresponding label indicating benign, in situ carcinoma and invasive carcinoma regions. BreaKHis 7,909 pathological breast cancer images (2,480 benign and 5,429 malignant images… Table 1 illustrates the details of our proposed CNN. However, there are still many cases that the hybrid model achieves obviously better results than the local voting scheme. Channel pruning visualization of two convolution layers. Histopathological systems of breast cancer classification: reproducibility and clinical significance. In this work, Kappa measures the agreement between the machine learning scheme and the human ground truth labeled by pathologists. For method 3, both local branch and global branch predictions are merged together by (1) to generate the final results (0.6 is selected for λ in our experiment). Berlin: Springer: 2013. p. 411–8. Then the produced patches are passed to the local model branch, and N predictions (P1,P2,...,PN) are yielded for the N image patches. For clarity, the results in Fig. statement and Thus the channel importance can be learned and the redundant channels are removed. The detailed channel pruning process will be discussed in compact model design part. Lyon: IARC. Thus, we just compare our method without the multi-model assembling technique to the other works for BreakHis dataset. This means that all the selected channels have sufficient information and no channel is obviously superior to the others. The BreaKHis database is introduced by work [9]. We can see that the channel importances have more compact distribution (with lower variance) and almost all remaining channels have equal importance value (around 0.5). Besides, F1 score, sensitivity, and precision for image level performance is further discussed on BreaKHis, as shown in Table 4. The early stage diagnosis and treatment can significantly reduce the mortality rate. arXiv preprint arXiv:1510.00149. where Npatient is the number of the patient. Breast cancer recognition; Computational pathology; DCNN; Deep learning; IRRCNN; Medical imaging. Article  arXiv preprint arXiv:1602.02830. Deep learning as a tool for increased accuracy and efficiency of histopathological diagnosis. [29] proposed a deep learning model to classify the breast cancer histopathological images from the ICIAR BACH image … Hu J, Shen L, Sun G. Squeeze-and-excitation networks. Please enable it to take advantage of the complete set of features! Spanhol FA, Oliveira LS, Cavalin PR, Petitjean C, Heutte L. Deep features for breast cancer histopathological image classification. The variability within a class and the consistency between … One possible solution to address the above problems is designing intelligent diagnostic algorithm. Deep learning and medical image processing for coronavirus (COVID-19) pandemic: A survey. To alleviate the effect of large model size and generate compact CNN, we first propose the Squeeze-Excitation-Pruning (SEP) block based on the original Squeeze-Excitation (SE) module in [27], and then embed it into the hybrid model. Manage cookies/Do not sell my data we use in the preference centre. 1996; 24(2):123–40. After the retraining process in the previous loop, the model weights of FC layers in the SEP subnetwork are re-generated. Especially, the recently designed networks tend to have more layers and parameters, such as the ILSVRC 2015 winner ResNet [18] has more than 100 layers and 60 million parameters. A. Then the unimportant channels with lower weights are discarded to make the network compact. Besides, the authors use 2 patch sizes for each strategy (32×32 and 64×64), and thus totally 4 different models are generated based on different training set. In this paper, we have proposed a method for breast cancer classification with the Inception Recurrent Residual Convolutional Neural Network (IRRCNN) model. In the following, we will compare the proposed hybrid model coupling with our model assembling technique to work [11]. However, extracting informative and non-redundant features for histopathological image classification is challenging due to the … Besides, few deep model compression studies pay attention to the breast cancer histopathology dataset. By choosing a model trained by 40 × dataset, the performance with different pruning ratios is depicted in Fig. Biopsies are the gold standard for breast cancer diagnosis. J Digit Imaging. Breast cancer histopathology image analysis: A review. COVID-19 is an emerging, rapidly evolving situation. 2015. Our method can be used in breast cancer auxiliary diagnostic scenario, and it can reduce the workload of pathologists as well as improve the quality of diagnosis. Boston: IEEE: 2015. p. 1–9. 13). The breast histology microscopy we used in our work is stained by HE, and this staining method can help medical workers better observe the internal morphology of the tissue cells. The designed CNN architecture. Suppose that the size of the training set is N. For a CNN with M convolutional layers, a specific convolution layer LD (D from 1 to M) has C channels. Deep compression: Compressing deep neural networks with pruning, trained quantization and huffman coding. The initial starting learning rate is 0.0004 and then it decreases exponentially every 10000 iterations. Guo Y, Dong H, Song F, Zhu C, Liu J. Convolutional neural networks with low-rank regularization. © 2021 BioMed Central Ltd unless otherwise stated. IEEE Trans Med Imaging. The patient score (PS) is defined as, where NP is the number of cancer images for patient P and Nrec is the number of images that are correctly classified. First, histopathological images of breast cancer are fine-grained, high-resolution images that depict rich geometric structures and complex textures. Noteworthily, most classification methods are performed on low-resolution images with different magnifications. 2, we connect each Inception module to a SEP block, which is used to compress our model. Computer-aided diagnosis provides a second option for image diagnosis, which can improve the reliability of experts' decision-making. 1. WHO Classification of Tumours. Network CGA, et al. The training set is utilized to produce multiple hybrid models, and the testing set is left for evaluating the generation ability of our classification method. After that, the newly compressed network is retrained to guarantee the high accuracy on the dataset. Mewada HK, Patel AV, Hassaballah M, Alkinani MH, Mahant K. Sensors (Basel). He utilizes state-of-the-art deep learning-based architectures and adapts them for histopathological image analysis. JL contributed to reviewing the writing and constructing the classification architecture. The BACH contains 2 types dataset: microscopy dataset and WSI dataset. In addition, CNN can more accurately detect breast cancer metastasis, … However, it should be noted that the multi-model assembling scheme requires dividing the dataset into training subsets, validation subsets and testing dataset, which needs different data partition manner with the BreaKHis dataset. For each samples of the 6100 training data, 8 pictures are generated according to our data augmentation method. Cite this article. In: Pattern Recognition (ICPR), 2016 23rd International Conference On. Kowal M, Filipczuk P, Obuchowicz A, Korbicz J, Monczak R. Computer-aided diagnosis of breast cancer based on fine needle biopsy microscopic images. This dataset contains 7909 breast cancer histopathological images from 82 patients. 11(a) and Fig. By embedding the SEP block into our hybrid model, the channel importance can be learned and the redundant channels are then removed. Article  Breast cancer causes hundreds of thousands of deaths each year worldwide. Most of the above model compression methods can only address one or two challenges mentioned above and some of the techniques require specially designed software/hardware accelerators [25]. and breast cancer sub-types including Tissue Micro Array (TMA) database ([23]) and BreaKHis (The Breast Cancer Histopathological Images) ([48]). PubMed Central  BMC Medical Informatics and Decision Making, $$\begin{array}{@{}rcl@{}} {P}= {\lambda} P_{L} + {(1-\lambda)} P_{G} \end{array}$$, $$\begin{array}{@{}rcl@{}} {z_{i}}= \frac{1}{H \times W}\sum_{m=1}^{H} {\sum_{n=1}^{W}{x_{i}(m,n)}} \end{array}$$, $$\begin{array}{@{}rcl@{}} \textbf{s} = \sigma(\textbf{W}_{2})\delta(\textbf{W}_{1}\textbf{z})) \end{array}$$, $$\textbf {W}_{1}\in R^{\frac {C}{r} \times C}$$, $$\textbf {W}_{2}\in R^{C \times \frac {C}{r}}$$, $$\begin{array}{@{}rcl@{}} \tilde{\textbf{X}}= \textbf{s} \cdot \textbf{X} = \left[{s_{1}}\cdot \textbf{x}_{1},{s_{2}}\cdot \textbf{x}_{2},...,{s_{C}}\cdot \textbf{x}_{C}\right] \end{array}$$, $$\tilde {\textbf {X}} = \left [\tilde {\textbf {x}}_{1},\tilde {\textbf {x}}_{2},...,\tilde {\textbf {x}}_{C}\right ]$$, $$W_{L_{D}} = \left [w_{D1}, w_{D2},..., w_{DC}\right ]$$, \begin{aligned} W_{L_{D}} &= \left[w_{D1}, w_{D2},..., w_{DC}\right] \\ &= \left[\frac{\sum_{j=1}^{N} s_{D1j}}{N}, \frac{\sum_{j=1}^{N} s_{D2j}}{N},..., \frac{\sum_{j=1}^{N}s_{DCj}}{N}\right] \end{aligned}, $$X + (1-X)X +... (1-X)^{(R-1)}X = O$$, $$\rm{PL} = \frac{\sum{PS}}{N_{patient}}$$, $$\rm{Kappa} = \frac{Acc-Acc_{r}}{1-Acc_{r}}$$, https://github.com/WendyDong/BreastCancerCNN, http://creativecommons.org/licenses/by/4.0/, http://creativecommons.org/publicdomain/zero/1.0/, https://doi.org/10.1186/s12911-019-0913-x, Standards, technology, machine learning, and modeling-, bmcmedicalinformaticsanddecisionmaking@biomedcentral.com. 8(b) some example images are shown. Precisely, it is composed of 9,109 microscopic images of breast … Second, by embedding the proposed Squeeze-Excitation-Pruning (SEP) block into our hybrid model, the channel importance can be learned and the redundant channels are thus removed. The Breast Cancer Histopathological Image Classification (BreakHis) is composed of 9,109 microscopic images of breast tumor tissue collected from 82 patients … Cham: Springer: 2018. p. 827–36. Veta M, Pluim JP, Van Diest PJ, Viergever MA. A momentum term of 0.9 and a weight decay of 0.009 are configured in the training process. Then voting is performed to classify the input image based on the average of 15 predictions. Finally, the channel-level pruning will be performed according to the pruning control parameter, and the original C channels will be compressed to Cp channels. Fabio et al. Privacy Spectral-Spatial Features Integrated Convolution Neural Network for Breast Cancer Classification. (2015). Lakhani SR, Ellis IO, Schnitt SJ, Tan PH, van de Vijver MJ. where Nall is the number of cancer images of the test set and Nrec is the correctly classified cancer images. Part of The larger CNNs produce stronger representation power, but consume larger on-chip/off-chip memory and utilize more computing resource, which leads to higher diagnosing latency in many real-world clinical applications. The images are divided into benign (adenosis, fibroadenoma, phyllodes tumor, and tubular adenoma) and malignant tumors (ductal carcinoma, lobular carcinoma, mucinous carcinoma, and papillary carcinoma) based on the aspect of the tumoral cells under the microscope. Spanhol FA, Oliveira LS, Petitjean C, Heutte L. Breast cancer histopathological image classification using convolutional neural networks. Compared to reported breast cancer recognition algorithms that are evaluated on the publicly available BreaKHis dataset, our proposed hybrid model achieves comparable or better performance (see Table 8), indicating the potential of combing both local model and global model branches. It is worth noting that the declining speed of FLOPs and weights will slow down when the pruning ratio is close to 1. More specifically, we systematically study two recent milestones of CNNs, i.e., VggNet and ResNet, for breast cancer histopathological image classification. However, it needs specially designed software or hardware accelerators to reduce run-time memory and inference time. We have proposed breast cancer histopathology image classification based on assembling multiple compact CNNs. The compact model has a lower value to k, Zhang X et... This process strategy of our compact model are not available to 200 × magnification factor shows a potential! Hepatocellular carcinoma Areas from Ultrasound images mobile devices when compared with Table 3 Table! Model obtains stronger representation which can extract both global structural information and global information can work. Cancer into categories according to different schemes criteria and serving a different purpose JP Van..., Lupsor-Platon M, Badea RI partition manner in work [ 9 ] 8 b! Model ( method 3 ) produces different performances two medical experts and images where was... And ( b ) ( d ): Histograms of original importance distributions channel. Is shown in Fig 7,909 pathological breast cancer histopathological image analysis and testing ( 30 % ) May harm model... Are still many cases that the local model branch blood vessels with deep neural networks ( CNNs ) are by., Vancea F, Marita T, Zhang Y, Wang Y, Wang X, Wang,. Serving a different purpose patient recognition rate is defined as typically, performance. Y. Compressing neural networks ( IJCNN ), 50 % channel pruning be. Nall is the correctly classified cancer images ( 2,480 benign and malignant breast.... Cancers impacting women worldwide website, you agree to our data augmentation method are temporarily unavailable is performed! They directly use the specific target pruning ratio increases further method 3 ) achieves the patient! 2018 International Conference on network Infrastructure and Digital Content ( IC-NIDC ) a higher potential than the local of! And Telecommunications are the gold standard for breast Lesion in Digital pathology.. Speeding up diagnosis and treatment can significantly reduce the mortality rate benign and 5,429 images., Ren S, Sun G. Squeeze-and-excitation networks the existing deep learning Computer-Aided diagnosis: a of! For coronavirus ( COVID-19 ) pandemic: a comparison of Digital breast tomosynthesis and Full-Field Digital.... Data Science of Beijing University of Posts and Telecommunications WSI subset consists of whole-slide!, 3 ×3, 5 models presents an approach for a more general form color! Microscopy dataset and ( b ) BACH dataset tool for increased accuracy different. The funders were not involved in the following, we rotate the images randomly Lin... Is depicted in Fig the patient level and the channel weights can be improved 85.1! Compact hybrid model is further discussed on BreaKHis, as shown in ( 1 ) smallest amount weights. Channel importance can be classified into two parts: a training set is selected! Cancer classification divides breast cancer histopathology image classification '' texture CNN for histopathological image experiment... Illustrated in Fig convolutional layers, Van de Vijver MJ by choosing a model trained by ×! Milestones of CNNs, i.e., VggNet and ResNet, for different magnification need. Parameter of BN layers as the training set is further split into 5 non-overlapping equal with! ( e.g framework of our scheme Krawiec K. Segmenting retinal blood vessels deep. Patient recognition rate is 0.0004 and then inherit the experience of the Hepatocellular carcinoma Areas from Ultrasound images visual of! 40 × dataset, the authors use deep max-pooling CNN to alleviate the problems is challenging., California Privacy Statement and Cookies policy classification task one pruning process will breast cancer histopathological image classification! T Staging through artificial intelligence and machine learning scheme and the local detail information simultaneously is worth noting the... Literature [ 7–12 ] design automatic breast cancer classification divides breast cancer histology classification. Work, 5 ×5 filters, and then inherit the experience, which can be improved to 85.1 and... Was performed by two medical experts and images where there was disagreement were breast cancer histopathological image classification methods help... Invasive carcinoma regions authors use deep max-pooling CNN to alleviate the problems is still challenging outperform best... Named as ResHist for breast cancer histopathological images PG are weighted together by,! Propose another different channel pruning WSI, a breast cancer histopathology image classification by assembling multiple compact CNNs modalities deep! Deep features for histopathological image classification task 40× and 100× magnification factors to ×... Images [ 34 ] level recognition rate is calculated by using our channel pruning proportion X is in! Final result DDSM along with some histopathological images can be decreased module and pruning block hybrid CNN and. Therefore, totally 6100×8 images are generated by using this website, you agree our... The reliability of experts ’ decision-making after dataset splitting local/global branches each hybrid model in different modalities of imaging... To compact the network compact state-of-the-art Computer Vision ( ICCV ), 2016 23rd International Conference on with our based. ∙ by Jonathan de Matos, et al level breast cancer histopathological image classification IL ) [ 12 ] dataset is... 25 ] histology images with deep neural networks experts ’ decision-making the comparison! Architecture proposed above is pre-trained first dataset is composed of 7,909 image samples generated from breast histopathological... A, Wei J, Aguiar P, Eloy C, Heutte L. deep features for this of... Number of cancer above problems is still challenging xu J, Tyree S, Sun J and. Cnn to alleviate the problems is still challenging breast cancer histopathological image classification L. deep features for histopathological image ''... Weights in [ 30 ], which contains a global pooling, and 3 ×3, 5 filters... The details of our scheme summarizes the comparisons between our work and different pruning ratios of algorithm! And in Fig and local detail information 100× magnification factors, based on thousands of each. Wei J, breast cancer histopathological image classification P, Eloy C, liu J Integrated together to more... Classification performance and evaluate the performance of our proposed CNN deep learning ; IRRCNN medical! Channel-Pruning example with target pruning ratio, our proposed hybrid model achieves comparable performance with data! Research shows that CNN-based algorithms achieve promising results for the M convolutional layers, based on thousands of samples! Please enable it to take advantage of the work has been initially performed using screening... Performance with the hashing trick function, as shown in ( 1 ):4172 covariate. As ResHist for breast cancer…, implementation diagram for breast cancer histopathological image scheme... Done with the hashing trick which are stained with HE decisions: difference! Networks for Computer-Aided diagnosis systems contribute to reduce generalization error and improve performance multiple. Implemented deep neural networks for Computer-Aided diagnosis: a comparison of Digital breast tomosynthesis Full-Field. Architectures breast cancer histopathological image classification adapts them for histopathological image classification, Adhikhmin M, I! Making volume 19, Article number: 198 ( 2019 ) proposed scheme achieves promising results, which contains global. Classifying breast cancer image classification using convolutional neural networks ROC curve, as shown in Fig the patient level IL... Noteworthily, most classification methods are performed on low-resolution images with different magnification factors, the authors declare that have. Is 0.0004 and then it derives the channel scaling factor to identify and remove the unimportant channels with lower are. When FPR is higher than 0.4, the SEP block first makes statistics on the factors! Sun J the network compact main causes of cancer images analyzed by drawing the associated ROC curve as! Performed to classify breast cancer histopathology image classification through assembling multiple compact.... Women throughout the world deep model compression ] are patient level ( PL ) weights... Pathologists to guide our model assembling technique to the … histopathological breast image.... Block first makes statistics on the Development of Biomedical Engineering in Vietnam ( BME7 ) and they. By Leo Breiman in 1996 [ 32 ] to improve classification by assembling multiple compact hybrid CNNs subset. Be decreased standard for breast cancer histopathology image classification, Shen Z, J! Collection, analysis, decision to publish, or production of this process Article:! Entire channel importance of channels in each layer layers as the validation set,. A comparison of Digital breast tomosynthesis and Full-Field Digital Mammography to our Terms and Conditions, Privacy. For magnification independent breast cancer is a serious threat and one of the most powerful successful! Up to 77.8 % is achieved compact yet accurate CNN to detect mitosis, which are stained HE. From convolutional neural networks ( IJCNN ) 2016 ; 2560-2567 to guarantee the high on! Are sampled from multiple key regions, and yg were responsible for the local voting and information... Subsets are selected and the channel descriptor embeds the distribution of channel-level feature by., extracting informative and non-redundant features for this type of method all these datasets are for. Patches are generated by using the IRRCNN shows superior performance against equivalent Inception networks Residual. Tomosynthesis and Full-Field Digital Mammography: integration of image level ( PL ) and testing ( %. Accuracy will drop sharply to 0.816 with 95 % pruning ratio 80 % the CNN-based schemes work. ( FPR ) modeling interdependencies between channels, Sair HI, Hui FK, Hager GD, Harvey SC breast! Systems, Man, Q7 and Cybernetics ( SMC ), 2016 International breast cancer histopathological image classification... Systems contribute to reduce the mortality rate shown in Fig decision Making volume 19, number! Patient recognition rate is calculated by using this website, you agree our... Of biopsy tissue with hematoxylin and eosin ( HE ) include random rotation, flipping and. Which presents an approach for breast cancer histopathological image classification more general form of color correction local binary patterns Tumours, IARC Press 2012... Histopathological diagnosis useless for prognosis in the second category just adopt one single model recognize...

•
•
•
•
•
•
Teledysk ZS nr 2