FOCUS: Knowledge-enhanced Adaptive Visual Compression for Few-shot Whole Slide Image Classification

TL;DR

FOCUS tackles few-shot WSI classification in computational pathology by integrating pathology foundation models with language priors to compress and prioritize diagnostically informative patches. It introduces a three-stage adaptive visual compression pipeline—global redundancy removal via FM features, language-guided patch prioritization, and sequential local compression—followed by cross-modal aggregation with pathology prompts for final prediction. Across TCGA-NSCLC, CAMELYON, and UBC-OCEAN, FOCUS achieves state-of-the-art or competitive performance in 4-, 8-, and 16-shot settings, with notable gains in the most resource-constrained 4-shot regime. Ablation studies confirm that each component and the choice of FM encoder and LLM prompts contribute meaningfully to the improvements, demonstrating effective transfer of pathological knowledge to data-limited WSI analysis.

Abstract

Few-shot learning presents a critical solution for cancer diagnosis in computational pathology (CPath), addressing fundamental limitations in data availability, particularly the scarcity of expert annotations and patient privacy constraints. A key challenge in this paradigm stems from the inherent disparity between the limited training set of whole slide images (WSIs) and the enormous number of contained patches, where a significant portion of these patches lacks diagnostically relevant information, potentially diluting the model's ability to learn and focus on critical diagnostic features. While recent works attempt to address this by incorporating additional knowledge, several crucial gaps hinder further progress: (1) despite the emergence of powerful pathology foundation models (FMs), their potential remains largely untapped, with most approaches limiting their use to basic feature extraction; (2) current language guidance mechanisms attempt to align text prompts with vast numbers of WSI patches all at once, struggling to leverage rich pathological semantic information. To this end, we introduce the knowledge-enhanced adaptive visual compression framework, dubbed FOCUS, which uniquely combines pathology FMs with language prior knowledge to enable a focused analysis of diagnostically relevant regions by prioritizing discriminative WSI patches. Our approach implements a progressive three-stage compression strategy: we first leverage FMs for global visual redundancy elimination, and integrate compressed features with language prompts for semantic relevance assessment, then perform neighbor-aware visual token filtering while preserving spatial coherence. Extensive experiments on pathological datasets spanning breast, lung, and ovarian cancers demonstrate its superior performance in few-shot pathology diagnosis. Codes are available at https://github.com/dddavid4real/FOCUS.

Figures (3)

• Figure 1: Illustration of the proposed FOCUS framework, which consists of three key components: (a) A knowledge-enhanced adaptive visual token compression module that performs global redundancy removal via FM features and language prior-guided visual token prioritization, (b) A sequential visual token compression module that performs pair-wise similarity thresholding to eliminate local redundancies while preserving spatial coherence, where $\sigma_i$ denotes the pairwise similarity (cosine similarity) between adjacent patches, and (c) A cross-modal aggregation module that combines the compressed visual features with pathology knowledge prompts for final prediction.
• Figure 2: Performance comparison of different foundation models (UNI, GPFM, Virchow, PLIP, and CONCH) on UBC-OCEAN under few-shot settings (measured by Balanced ACC).
• Figure 3: Performance comparison of prompts from different LLMs on UBC-OCEAN under 16-shot settings (Balanced ACC).