Computer-aided Pathological Annotation Framework: A Deep Learning-Based Diagnostic Algorithm of Lung Cancer

Whole-slide image (WSI) annotation is fundamental to computational pathology. However, pixel-level manual annotation of large-scale histopathology dataset is impractical due to the combination of required labor-intensitive tasks and the budget limitation in pathologists' supports. In this paper, we develop a framework for computer-aided annotation of cancerous region on WSI with pathologically conformed lung cancers aiming to significantly reduce the workload of pathologists in annotation on the WSIs. We used deep convolutional neural network (DCNN) to automate the labeling of the area of cancerous lesions. Our method was able to identify malignant area on tested WSIs with a mean dice of 0.84, in comparison to pathologists' manual annotation.