Classification of Histopathology Whole Slide Images Based on Self-supervised Learning and Multi-scale Feature Fusion

Histopathological whole slide images (WSI) usually contain rich biological information and are thus used as the gold standard for cancer diagnosis. However, due to problems such as inconspicuous morphological features of tissue and lack of pixel-level labels, the accuracy of WSI classification tasks using deep learning technology is low, especially in predicting recurrence. To solve this problem, we propose a new classification method based on self-supervised learning and multi-scale feature fusion, which utilizes contrastive learning to enhance feature representation and fuses multi-scale features of WSI to improve the accuracy of the predicting recurrence task. The proposed algorithm achieved the best results on a gastric cancer WSI dataset collected clinically, with an accuracy rate of $68.18 \%$, a recall rate of $70.00 \%$, and an F1 score of $66.67 \%$. These results show that our algorithm can effectively improve the classification accuracy, making it a promising method for computer-aided WSI classification.