基於深度學習於乳癌異常細胞之自動分類研究

Research of Automatic Breast Canner Atypia Cell Classification using Deep Learning Architectures

簡豪志(Hao-Jhih Jian)

指導教授 ： 張元翔

中原大學/電機資訊學院/資訊工程研究所/碩士(2017年)

https://doi.org/10.6840/cycu201700483

摘要

摘要
乳癌一直是致死率榜上有名的疾病之一，為了診斷與治療乳癌有一系列的程序，其中一項工作分析細胞切片影像是判定是否乳癌的關鍵，這仰賴人力去判斷分析且會因為每位醫師的判斷標準不同可能導致誤判。為此我們希望透過自動化分類的方式來提升效率與一個穩定判斷依據，針對乳癌切片細胞進行監督式機器學習之預測分類研究。使用資料為MITOS-ATYPIA-14比賽提供的訓練集與測試集。使用深度學習(Deep Learning)中的卷積神經網路(Convolutional Neural Networks)做影像特徵的提取、篩選及分類，搭配支援向量機(Support Vector Machine)來更進一步地對資料做分類，使用五種不同的組合方法分別為CNN Single-Model、CNN Multi-Model、CNN Multi-Model with SVM、CNN Probability Map with SVM、CNN Probability Map and Criteria Label with SVM來完成分類器，輸入一張乳癌切片顯為影像後分類器會產生1~3得分分別代表不同類別(低、中與高風險的異常細胞分布狀況)，實驗結果顯示與其他隊伍相比本研究提出的系統為較有前途的結果。最後，本論文系統能提供第二個意見(判斷依據)給病理學家在分析時當作參考。
關鍵詞：乳癌; 電腦視覺; 卷積神經網路; 深度學習.

關鍵字

乳癌 ； 電腦視覺 ； 卷積神經網路 ； 深度學習

並列摘要

Abstract
Breast cancer has been one of the major causes of death around the world. The diagnosis of breast cancer relies on the interpretation of microscopy images from breast biopsies by expert pathologists. In this study, we explored the feasibility of developing convolutional neural networks (CNNs) with deep learning architectures to automatically classify these images. The Mitos & Atypia 14 (MITOS-ATYPIA-14) contest database was used as the benchmark for training and testing. A scoring system with three different architectures, namely CNN Single-Model, CNN Multi-Model, CNN Multi-Model with SVM, CNN Probability Map with SVM, and CNN Probability Map and Criteria Label with SVM are proposed. Given a microscopy image, our system can be used to yield a score in the range of 1 ~ 3 representing different classes (i.e., low, moderate, and high risk of atypia cell malignancy). Experimental results demonstrate that our proposed systems have achieved promising results when compared with previous systems presented by other competitors. Ultimately, our system could be used as a “second opinion” to pathologists during the diagnosis of microscopy images from breast biopsies.
Keywords- Breast Cancer; Computer Vision; Convolutional Neural Networks; Deep Learning.

並列關鍵字

Breast Cancer ； Computer Vision ； Convolutional Neural Networks ； Deep Learning

參考文獻

[1] World Health Organization Media centre Face sheet “Cancer”, Feb. 2017.

[2] American Cancer Society “Cancer Facts and Figures 2017”.

[3] Health Promotion Administration, Ministry of health and Welfare, Taiwan, “Cancer registry annual report 2013”, Jan. 2016.

[4] A. M. Leitch, G. D. Dodd, M. Costanza, M. Linver, P. Pressman, L. McGinnis, and R. A. Smith, “American Cancer Society guidelines for the early detection of breast cancer: update 1997,” CA: A cancer Journal for Clinicians, 47(3), pp. 150-153, 1997

[5] N. Bayramoglu, J. Kannala, J. Heikkilä, “Deep learning for magnification independent breast cancer histopathology image classification” IEEE International Conference on Pattern Recognition (ICPR), pp. 2440-2445, 2016