Multi-Scale Double-Layer Multiple Instance Learning Algorithm for Whole Slide Image Classification of Primary Gastric Lymphoma

Primary Gastric Lymphoma (PGL) is a type of malignant lymphoma that originates in the stomach. Timely and precise diagnosis and classification are crucial for effective treatment and predicting outcomes. Advances in artificial intelligence, especially deep learning, have ushered in a new era for medical diagnostics. Weakly supervised learning, in particular, is gaining traction for classifying histopathological whole slide images (WSIs). Nonetheless, there remain challenges, such as the classification of small sample cohorts. To address these issues, we propose a novel multi-scale double-layer multiple instance learning classification algorithm for accurate classification of PGL. This algorithm leverages pseudo-bags-a technique to effectively augment the sample size-and employs multiple instance learning to distill essential features. It then constructs an enhanced classification model utilizing a multi-scale, double-layer approach. Validation using clinical pathological data indicates that our method outperforms other latest methods in classifying primary gastric lymphoma, as well as obtains superior results on the CAMELYON-16 dataset. Our framework holds promise for broader adoption across various medical applications.