A multi-task deep learning framework for perineural invasion recognition in gastric cancer whole slide images

Abstract

Perineural invasion (PNI) is the process of neoplastic invasion of the nerves and is a prognostic factor in gastric cancer. However, such examination is a labor-intensive task in high-resolution digital pathological images. To alleviate this problem, we propose a multi-task deep learning-based framework to highlight diagnostically significant PNI regions in whole slide images (WSIs) of human gastric cancer tissue sections. The proposed framework includes a gastric cancer segmentation model, neural detection model, and PNI decision-making module, which realizes the segmentation of the gastric cancer region while completing the task of identifying PNI. Adequate comparative experiments were performed on our own gastric cancer PNI dataset called GC-PNI. Experiments have shown that our proposed model can effectively diagnose PNI with a high sensitivity of 0.972 and a specificity of 0.933, illustrating the potential of this practical application.