Deep Learning for Identifying Metastatic Breast Cancer识别转移性乳腺癌_论文笔记

摘抄：

1.Standardized, accurate and reproducible pathological diagnoses are essential
for advancing precision medicine.

2.Limitations of the qualitative visual analysis of microscopic images includes lack of
standardization, diagnostic errors, and the significant cognitive load required to manually evaluate millions of cells across hundreds of slides in a typical pathologist’s workday.在一个典型的病理学家的工作日里，手动评估数百张幻灯片上的数百万个细胞所需要的巨大认知负荷.

3.Finally, combining the predictions of our deep learning system with a pathologist’s interpretations produced a significant reduction in the pathologist’s error rate.

4.图像预处理：采用基于阈值的分割方法来自动检测背景区域，生成掩码图像：In particular, we first transfer the original image from the RGB color space to the HSV color space, then the optimal threshold values in each channel are computed using the Otsu algorithm , and the final mask images are generated by combining the masks from H andS channels. 

5. 切片级别的评估：AUC分数和整张切片包含癌症的预测可能性。Competition participants submitted a probability for each test slide indicating its predicted likelihood of containing cancer. The competition organizers measured the participant performance using the area under the receiver operator (AUC) score.

框架：

patch级别的标注：如果一个patch块位于肿瘤区域，则为肿瘤/阳性patch，标记为1，否则为正常/阴性paych，标记为0。

在实验中，评估了40倍、20倍和10倍的放大倍数，得到了40倍放大倍率下的最佳性能。

热图：

在完成基于patch的分类阶段后，为每个WSI生成肿瘤概率热图。在这些热图上，每个像素包含一个0到1之间的值，表示像素包含肿瘤的概率。

对于基于幻灯片的分类任务，后处理将每个WSI的热图作为输入，并将整个WSI的单个肿瘤概率作为输出。给定热图，从每张热图中提取28个几何和形态特征，包括肿瘤区域占整个组织区域的百分比，肿瘤区域与周围最小凸区域的面积比，平均预测值，肿瘤区域的最长轴。计算了所有训练案例中肿瘤概率热图上的这些特征，并建立了一个随机森林分类器来区分有转移的阳性wsi和阴性wsi。