應用機器學習之電腦輔助系統於病理組織影像的分析及診斷

Computer-aided Analysis and Classification in Histopathology Images using Machine Learning

陳巧旻(Chiao-Min Chen)

指導教授 ： 張瑞峰

國立臺灣大學/電機資訊學院/資訊工程學研究所/博士(2020年)

https://doi.org/10.6342/NTU202000690

摘要

病理學是病患照護與治療的理論基礎。做為病理學的疾病研究輔助，病理組織影像已經被使用來評估組織樣本中的可疑變異，為了要建立更有效率的診斷程序，許多的電腦輔助診斷系統已經被研發做為輔助醫師之工具。目前，前列腺癌已經成為男性中具性命威脅的疾病。對於前列腺癌來說，一般常規染色的組織切片影像是廣泛用於臨床檢查與術後評估。因此在這篇論文研究中，藉由電腦輔助診斷系統量化組織學影像特徵，透過萃取良性與惡性腺體之特性用於區分良性、惡性腺體和其他組織，進一步提供惡性腺體分佈之結果。由於組織學結構分析是病理學家用於評估治療方針之方式，包括格里森分級和分級檢測。為此本研究進一步的使用深度學習技術用於組織學結構分析與檢測惡性組織分佈之關係，以提供病理學家便利與輔助診斷的協助。為了進一步解析疾病的發展，蛋白質組拓撲對於研究蛋白質的空間調控以及蛋白質與蛋白質的相互作用非常重要。因此開發影像套合系統，可以將不同的局部染色(免疫組織化學染色)的連續切片進行重新對齊並校正組織細胞內的局部偏差以重新獲得空間對應關係。因此，可以篩選蛋白質與蛋白質相互作用以建立蛋白質組拓撲提供醫師做為解析疾病發展之輔助工具。

關鍵字

前列腺癌 ； 病理學 ； 電腦輔助診斷 ； 機器學習 ； 深度學習 ； 影像套合

並列摘要

Pathology is a significant field in modern medical diagnosis for disease research. As an adjunct to pathological disease research, a digital pathology image has been used for evaluating suspicious abnormalities in the tissue specimen. In order to establish efficient diagnostic procedures, various computer-aided diagnosis (CAD) systems have been developed to assist pathologists. Prostate cancer has become a major health concern in aging males. For prostate cancer, whole-slide conventional stained tissue section slices are the most common imaging and are widely used on clinical examination and postoperative evaluation. In this study, the development of CAD system based histological characteristics can provide the result of malignant gland distribution. For the histological structure analysis, the CAD system is presented with several improvements. Consequently, the development of the CAD system based on deep learning and spatial statistics are increased to delineate malignancy regions. Furthermore, proteome topology is necessary for studying the spatial regulation of proteins and protein-protein interactions for the prediction of actual disease progression. Hence, an effective image registration algorithm to minimize position deviation in a serial overlays of whole-slide immunohistochemical (IHC) images. By image registration, different IHC-stained sections could be digitized, realigned and corrected local deformation to regain spatial correspondence. Protein-protein interaction can be thus screened to establish the proteome topology for providing pathologists with convenience and assist in analyzing disease development.

並列關鍵字

Prostate cancer ； pathology ； computer-aided diagnosis ； machine learning ； deep learning ； image registration

參考文獻

[1] A. M. Marchevsky and M. R. Wick, "Evidence-based medicine, medical decision analysis, and pathology," Human pathology, vol. 35, no. 10, pp. 1179-1188, 2004.

[2] C. Cui, S. Fan, H. Lei, X. Qu, and D. Zheng, "Deep learning-based research on the influence of training data size for breast cancer pathology detection," The Journal of Engineering, vol. 2019, no. 23, pp. 8729-8732, 2019.

[3] M. Y. Gabril and G. M. Yousef, "Informatics for practicing anatomical pathologists: marking a new era in pathology practice," Modern Pathology, vol. 23, no. 3, pp. 349-358, 2010.

[4] X. Li and K. N. Plataniotis, "A complete color normalization approach to histopathology images using color cues computed from saturation-weighted statistics," IEEE Transactions on Biomedical Engineering, vol. 62, no. 7, pp. 1862-1873, 2015.

[5] A. M. Khan, N. Rajpoot, D. Treanor, and D. Magee, "A nonlinear mapping approach to stain normalization in digital histopathology images using image-specific color deconvolution," IEEE Transactions on Biomedical Engineering, vol. 61, no. 6, pp. 1729-1738, 2014.