面向WSI和MSI病理图像的通用型癌细胞检测与计数方法研究

病理检查是疾病诊断的金标准，有着相比临床诊断更高的准确率。病理图像上的癌细胞检测和计数是各项癌症诊断的基石。整组织切片扫描图像(WSI)和光学显微镜图像(MSI)是两种最常用的病理图像。现有方法单纯在WSI数据上进行癌细胞检测和计数，难以在WSI和MSI上通用。然而由于WSI制片周期长，设备不易随时取用，临床上审片依然主要依赖MSI数据，这极大地限制了现有方法的应用前景。针对跨WSI和MSI通用型癌细胞检测和计数，这一紧贴临床实际需求的问题，目前尚无有效解决方案。本课题针对跨域病理图像理解模型构建、密集环境下癌细胞高精度检测、癌细胞自适应计数等内容开展深入研究。旨在：1）从单纯WSI内部图像理解到面向WSI和MSI跨域通用型感知，让病理图像分析“紧贴需求”；2）从像素空间小目标检测到编码空间信号拟合，让癌细胞检测“不惧密集”；3）从单纯密度图估计到双重自适应结构，让癌细胞计数“灵活精准”。

Pathological examination is the gold standard of disease diagnosis, which has higher accuracy than clinical diagnosis. The detection and counting of cancer cells on pathological images is the cornerstone of cancer diagnosis. At present, Whole Slide Images (WSI) and Microscopy Images (MSI) are the two fundamental pathological analysis data types. Existing pathological image analysis methods aim at WSI, and are difficult to be applied in both kinds of data. However, it is not easy to access the WSI equipment at any time, and the production period of WSI data is quite long. Pathologists still rely on MSI data in their daily examination. This greatly limits the application of the existing methods in the clinical diagnosis. Furthermore, there is no effective solution to the problem of universal cancer cell detection and counting across WSI and MSI, which is a problem close to the needs in clinical practices. This project plans to conduct in-depth research on construction of cross domain pathological image understanding model, high precision detection of cancer cells and adaptive stable counting of cancer cells. The purpose is: 1) from simple WSI internal image understanding to WSI and MSI cross domain universal perception, making pathological image analysis "close to demand"; 2) from small target detection in pixel space to signal regression in encoding space, making cancer cell detection "not afraid of high density"; 3) from simple density map estimation to dual adaptive structures, making cancer cell count "flexible and accurate".