Semantics-Aware Attention Guidance for Diagnosing Whole Slide Images

arXiv:2404.10894v1 Announce Type: new
Abstract: Accurate cancer diagnosis remains a critical challenge in digital pathology, largely due to the gigapixel size and complex spatial relationships present in whole slide images. Traditional multiple instance learning (MIL) methods often struggle with these intricacies, especially in preserving the necessary context for accurate diagnosis. In response, we introduce a novel framework named Semantics-Aware Attention Guidance (SAG), which includes 1) a technique for converting diagnostically relevant entities into attention signals, and 2) a flexible attention …

abstract arxiv attention cancer cancer diagnosis challenge context cs.cv diagnosis digital digital pathology guidance images instance mil multiple pathology relationships semantics spatial struggle type