基于WSI的病理图像辅助诊断技术研究

随着数字病理和机器学习技术的深入发展，基于数字全切片图像（WSI）的癌症辅助诊断方法成为计算机辅助诊断领域的研究热点。本课题从医学临床诊断实际需求出发，以国家卫生计生委癌症诊断质量控制平台的病理全切片图像病例为数据基础，重点开展乳腺、消化系统、妇产科等癌症的计算机辅助诊断方法研究。在研究解决数字病理全切片图像归一化、诊断相关区域检测与分割、病理学知识与深度学习相结合的病理特征提取相关问题基础上，研发出基于诊断相关区域检索的辅助诊断方法和基于弱监督学习的病理全切片辅助诊断方法。为诊断提供病变关键区域定位、相关典型历史病例等辅助信息，并进行癌变类型识别，在乳腺、消化道、妇产科数据集上肿瘤良恶性的分类精度达到95%以上。该研究是图像处理、模式识别、医学图像分析领域的交叉与难点问题，研究成果将对提高我国医学病理整体诊断水平、加强病理诊断质量控制有重要的理论意义和实际应用价值。

Owing to the development of digital pathology and machine learning, computer-aid-diagnosis (CAD) methods for cancer via whole-slide-image (WSI) has become a research focus. In this project, CAD approach for breast, digestive and gynecological cancer will be developed based on digital cases and WSIs, which are supplied by the Quality Control Platform of National Health and Family Planning Commission of PRC. Specifically, the approaches of WSI normalization, diagnostically relevant region segmentation and pathobiology-knowledge-guided feature learning are first researched. Then, CAD methods based on diagnostically relevant region retrieval and weakly supervised learning will be developed. The aim of these CAD methods is to provide the location of lesion region, the prediction of cancer categories and relevant cases diagnosed by pathologists, reaching an accuracy of 95% in diagnosis of breast, digestive and gynecological cancer. In summary, this research is an interdisciplinary and challenging aspect of image processing, pattern recognition and medical image analysis, and is theoretically and practically significant for the improvement of diagnostic level and quality control in cancer diagnosis.