Local Attention Graph-Based Transformer for Multi-target Genetic Alteration Prediction

Classical multiple instance learning (MIL) methods are often based on the identical and independent distributed assumption between instances, hence neglecting potentially rich contextual information beyond individual entities. On other hand, Transformers with global self-attention modules have been proposed to model interdependencies among all instances. However, in this paper we question: Is relation modeling using necessary, or can appropriately restrict calculations local regimes large-scale whole slide images (WSIs)? We propose a general-purpose attention graph-based Transformer for MIL (LA-MIL), introducing an inductive bias by explicitly contextualizing instances adaptive of arbitrary size. Additionally, efficiently adapted loss function enables our approach learn expressive WSI embeddings joint analysis biomarkers. demonstrate that LA-MIL achieves state-of-the-art results mutation prediction gastrointestinal cancer, outperforming existing models important biomarkers such as microsatellite instability colorectal cancer. Our findings suggest sufficiently dependencies par modules. implementation is available at https://github.com/agentdr1/LA_MIL .