對於病理分析來說，肝臟腫瘤及其壞死區域的偵測是必要的。在肝臟組織中，當腫瘤死因於缺氧或是淋巴球的攻擊時就會有壞死的出現。傳統上病理檢測是由專業的病理醫師進行的，此過程耗時且容易因為疲勞導致錯誤，因此急切地需要電腦輔助診斷的方法。為了這個需求，這篇論文提出了一個嶄新的模型，名為多尺度強化模型，來在肝臟病理切片中偵測腫瘤及壞死區域。對比單倍率小影像的模型，此篇提出的模型更能模擬實際醫師診斷過程，以各種尺度去分析影像。此神經網路的架構由三個部分組成。第一部分從不同尺度的影像提取特徵，第二部分是將上一步驟的特徵整合，最後則是預測出腫瘤及壞死區域。另一方面，在醫學影像的標註成本是很昂貴的，因為仰賴專業的病理知識。此外，非腫瘤區域是多於腫瘤區域的，而腫瘤區域又更多於壞死區域，因此此篇論文提出一個優化的主動學習演算法自適應挑選少數樣本，可以解決資料不平衡的問題及提升在有限標註下的偵測效果。此篇論文的模型採用敏感度及交集聯集比進行評比，在腫瘤及壞死敏感度上可以達到平均95.4%和95.5%，在腫瘤及壞死IoU上可以達到平均91.3%和94.3%。此外，自適應主動學習演算法在有限標註偵測的準確度比普通主動學習方法及隨機採樣來的好。此論文的貢獻有以下三個，(1) 我們提出一個嶄新的模型偵測腫瘤及壞死區域，並提供一個客觀的指標。(2) 我們設計了一個主動學習的演算法用於病理組織偵測。(3) 我們發展出一個採用自適應主動學習的方法解決採樣資料不平衡。

Liver tumor and necrosis detection are essential steps for pathologic analysis. In liver tissue, necrosis appears when the tumor cells die as the result of hypoxia or lymphocyte attack. Pathologic examinations are traditionally performed by professional pathologists, which are therefore time-consuming and prone to subjective errors. Therefore, automatic computer-aided examination methods for biomedical images are urgently required. To meet this requirement, the present study proposes a novel multi-magnification structure, referred to as multi-magnification attention convolutional neural network (MMA-CNN), to detect tumor and necrosis areas in liver stained WSIs. In contrast to single patch-wise CNNs, MMA-CNN more-closely mimics the actual working process of pathologists in employing a multi-magnification approach to analyze the images. The architecture of the network is composed of three parts. The first part is feature extraction from different magnification images. The next part is feature integration from the last part. Finally, predict and capture the regions of tumor and necrosis. On the other hand, the annotation cost in medical image is expensive because the annotations rely on professional knowledge. Besides, in general, the distributing area of normal is larger than tumor, while the distributing area of tumor is much larger than necrosis. This paper therefore proposes an improved active learning algorithm, which adaptively samples more minority in each iteration, to resolve the resulting data imbalance problem and enhance the effect of detection on the limited annotation. The performance of the proposed MAA-CNN is evaluated using the intersection over union (IoU) and sensitivity metrics. The result reaches the tumor and necrosis sensitivity with 95.4% and 95.5% in average, and the tumor and necrosis IoU with 91.3% and 94.3% in average, respectively. Besides, the performance of active learning with adaptive minority sampling achieve the better performance than random selection and active learning without adaptive minority sampling approach on the limited annotation. The contribution of this study has three-fold, (1) we proposed a novel network to detect tumor and necrosis regions, and provided an objective index. (2) we design an algorithm of active learning applying pathological tissue detection. (3) we develop an active learning approach with adaptive minority sampling to resolve the data imbalance in the selecting stage.

[1] F. Bray, J. Ferlay, I. Soerjomataram, RL. Siegel, LA. Torre, A. Jemal, “Global cancer statistics 2018: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries”
[2] L. C. Chen, G. Papandreou, I. Kokkinos, K. Murphy, and A. L. Yuille, “DeepLab: Semantic image segmentation with deep convolutional nets, atrous convolution, and fully connected CRFs,” IEEE Trans. Pattern Anal. Mach., 2018.
[3] K. Fukushima and S. Miyake, “Neocognitron: A self-organizing neural network model for a mechanism of visual pattern recognition,” in Competition and Cooperation in Neural Nets, 1982
[4] A. Waibel, T. Hanazawa, G. Hinton, K. Shikano and K. J. Lang, "Phoneme recognition using time-delay neural networks," in IEEE Transactions on Acoustics, Speech, and Signal Processing, 1989.
[5] Y. Lecun, L. Bottou, Y. Bengio and P. Haffner, "Gradient-based learning applied to document recognition," in Proceedings of the IEEE, 1998.
[6] A. Krizhevsky, I. Sutskever, G. Hinton, “ImageNet Classification with Deep Convolutional Neural Networks.” In Neural Information Processing Systems, 2012.
[7] K. Simonyan, A. Zisserman, “Very Deep Convolutional Networks for Large-Scale Image Recognition”, 2014.
[8] C. Szegedy et al., "Going deeper with convolutions," IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2015.
[9] K. He, X. Zhang, S. Ren and J. Sun, "Deep Residual Learning for Image Recognition," IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2016.
[10] G. Huang, Z. Liu, L. Van Der Maaten and K. Q. Weinberger, "Densely Connected Convolutional Networks," IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2017.
[11] J. Hu, L. Shen and G. Sun, "Squeeze-and-Excitation Networks," 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2018.
[12] V. Nair and G. E. Hinton, “Rectified linear units improve restricted boltzmann machines”, In Proceedings of the 27th International Conference on International Conference on Machine Learning (ICML), 2010.
[13] D. Misra, “Mish: A Self Regularized Non-Monotonic Neural Activation Function”, ArXiv, 2019.
[14] N. Srivastava, G. Hinton, A. Krizhevsky, I. Sutskever, and R. Salakhutdinov. “Dropout: a simple way to prevent neural networks from overfitting” J. Mach. Learn. Res. 15, 1, 2014
[15] A. Choromanska et al., “Batch normalization: Accelerating deep network training by reducing internal covariate shift,” arXiv Prepr. arXiv1502.03167, 2015.
[16] J. Duchi, E. Hazan, Y. Singer, “Adaptive Subgradient Methods for Online Learning and Stochastic Optimization”, J. Mach. Learn. Res., 2011
[17] T. Tieleman, G. E. Hinton, N. Srivastava, and K. Swersky, “Lecture 6.5-rmsprop: Divide the gradient by a running average of its recent magnitude,” COURSERA Neural Networks Mach. Learn., 2012.
[18] D. P. Kingma and J. L. Ba, “Adam: a method for stochastic optimization,” in International Conference on Learning Representations, 2015.
[19] T. A.M Elhassan, “Classification of Imbalance Data using Tomek Link(T-Link) Combined with Random Under-sampling (RUS) as a Data Reduction Method”, Journal of Informatics and Data Mining, 2016.
[20] S. Petushi, F.U. Garcia, M.M. Haber, C. Katsinis, and A. Tozeren, “Large-scale computations on histology images reveal grade-differentiating parameters for breast cancer.” BMC Med Imaging, 6, 14, 2006.
[21] S. Naik, S. Doyle, S. Agner, A. Madabhushi, M. Feldman, J. Tomaszewski, “Automated gland and nuclei segmentation for grading prostate and breast cancer histopathology.” Proceedings of IEEE ISBI, 2008.
[22] S. Doyle, S. Agner, A. Madabhushi, M. Feldman and J. Tomaszewski, "Automated grading of breast cancer histopathology using spectral clustering with textural and architectural image features," 2008 5th IEEE International Symposium on Biomedical Imaging, 2008.
[23] Po-Whei Huang and Y.H. Lai., “Effective segmentation and classification for HCC biopsy images.” Pattern Recognition., 2009.
[24] Y. Zhou, H. Chang, K. Barner, P. Spellman and B. Parvin, "Classification of Histology Sections via Multispectral Convolutional Sparse Coding," 2014 IEEE Conference on Computer Vision and Pattern Recognition, Columbus, 2014.
[25] A. Cruz-Roa, H. Gilmore, A. Basavanhally et al., “Accurate and reproducible invasive breast cancer detection in whole-slide images: A Deep Learning approach for quantifying tumor extent.”, 2017.
[26] L. Hou, D. Samaras, T. M. Kurc, Y. Gao, J. E. Davis, and J. H. Saltz, “Patch-based convolutional neural network for whole slide tissue image classification,” in
Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2015.
[27] W. Huang, “Automatic HCC Detection Using Convolutional Network with Multi-Magnification Input Images”
[28] H. Tokunaga, Y. Teramoto, A. Yoshizawa, R. Bise, “Adaptive Weighting Multi-Field-Of-View CNN for Semantic Segmentation in Pathology.” CVPR, 2019.
[29] Y. Xu, Z. Jia, Y. Ai, F. Zhang, M. Lai and E. I. Chang, "Deep convolutional activation features for large scale Brain Tumor histopathology image classification and segmentation," 2015 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), Brisbane, QLD, 2015
[30] N. V. Chawla, K. W. Bowyer, L. O. Hall, and W. P. Kegelmeyer, “SMOTE: synthetic minority over-sampling technique.” J. Artif. Int. Res. 16, 2002.
[31] T. Lin, P. Goyal, R. Girshick, K. He and P. Dollár, "Focal Loss for Dense Object Detection," 2017 IEEE International Conference on Computer Vision (ICCV), Venice, 2017.
[32] Z. Zhou, J. Shin, L. Zhang, S. Gurudu, M. Gotway, J. Liang, “Fine-Tuning Convolutional Neural Networks for Biomedical Image Analysis: Actively and Incrementally.” In Proceedings. IEEE Computer Society Conference on Computer Vision and Pattern Recognition, 2017.
[33] K. Wang, D. Zhang, Y. Li, R. Zhang, L. Lin, “Cost-Effective Active Learning for Deep Image Classification” IEEE Transactions on Circuits and Systems for Video Technology, 2016.
[34] D. Yoo and I. S. Kweon (2019), “Learning loss for active learning.” In IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2019.
[35] S. Ertekin, J. Huang, L. Bottou, C. Giles, “Learning on the border: Active learning in imbalanced data classification.” International Conference on Information and Knowledge Management, Proceedings. 2007.
[36] Chao-Ting Li, “Annotation-Effective Active Learning for Extreme Class Imbalance Problem: Application of Lymphocyte Detection in H&E Stained Liver Histopathological Image.”
[37] Sarker, Md. Mostafa Kamal & Rashwan, Hatem & Talavera, Estefanía & Banu, Syeda Furruka & Radeva, Petia & Puig, Domenec. “MACNet: Multi-scale Atrous Convolution Networks for Food Places Classification in Egocentric Photo-streams.”, 2018.
[38] https://medium.com/mlreview/a-guide-to-receptive-field-arithmetic-for-convolutional-neural-networks-e0f514068807
[39] T. Sheng, C. Feng, S. Zhuo, X. Zhang, L. Shen and M. Aleksic, “A Quantization-Friendly Separable Convolution for MobileNets," 2018 1st Workshop on Energy Efficient Machine Learning and Cognitive Computing for Embedded Applications (EMC2), Williamsburg, VA, 2018
[40] Razvan Pascanu, Tomas Mikolov, and Yoshua Bengio. “On the difficulty of training recurrent neural networks.” In Proceedings of the 30th International Conference on International Conference on Machine Learning (ICML'13), 2013.
[41] K. Nazeri, A. Aminpour, M. Ebrahimi, “Two-Stage Convolutional Neural Network for Breast Cancer Histology Image Classification.”, 2018