MiCo: Multiple Instance Learning with Context-Aware Clustering for Whole Slide Image Analysis

TL;DR

MiCo tackles spatial heterogeneity in histopathology MIL by introducing context-aware clustering with semantic anchors $S=\{s_k\}_{k=1}^K$ and patch features $h_m \in \mathbb{R}^d$. It deploys multi-layer modules consisting of Cluster Route and Cluster Reducer to dynamically link distant tissue patches of the same type and to consolidate redundant anchors, yielding refined instance representations. Across nine cancer datasets for survival and subtyping, MiCo achieves state-of-the-art performance (e.g., mean C-index $0.680$ and high ACC/AUC), with ablations confirming the necessity of anchors, CluRoute, and CluReducer. The approach enhances prognostic and diagnostic capabilities in WSI MIL and provides interpretable semantic anchors, with code released at the authors’ GitHub repository for reproducibility and further research.

Abstract

Multiple instance learning (MIL) has shown significant promise in histopathology whole slide image (WSI) analysis for cancer diagnosis and prognosis. However, the inherent spatial heterogeneity of WSIs presents critical challenges, as morphologically similar tissue types are often dispersed across distant anatomical regions. Conventional MIL methods struggle to model these scattered tissue distributions and capture cross-regional spatial interactions effectively. To address these limitations, we propose a novel Multiple instance learning framework with Context-Aware Clustering (MiCo), designed to enhance cross-regional intra-tissue correlations and strengthen inter-tissue semantic associations in WSIs. MiCo begins by clustering instances to distill discriminative morphological patterns, with cluster centroids serving as semantic anchors. To enhance cross-regional intra-tissue correlations, MiCo employs a Cluster Route module, which dynamically links instances of the same tissue type across distant regions via feature similarity. These semantic anchors act as contextual hubs, propagating semantic relationships to refine instance-level representations. To eliminate semantic fragmentation and strengthen inter-tissue semantic associations, MiCo integrates a Cluster Reducer module, which consolidates redundant anchors while enhancing information exchange between distinct semantic groups. Extensive experiments on two challenging tasks across nine large-scale public cancer datasets demonstrate the effectiveness of MiCo, showcasing its superiority over state-of-the-art methods. The code is available at https://github.com/junjianli106/MiCo.

Figures (3)

• Figure 1: Overview of MiCo. MiCo consists of multi-layered context-aware clustering modules. Each module is organized by a Cluster Route module, which aggregates and propagates semantic information, and a Cluster Reducer module, which consolidates redundant anchors while enhancing information exchange between semantically distinct anchors.
• Figure 2: A. Qualitative results of cluster assignment, showcasing WSI thumbnails, cluster assignment results, and corresponding regions of interest (ROIs) at different stages. Blue outlines denote ground truth tumor regions, while areas assigned to the same cluster are displayed in identical colors. B. Interpretability analysis of MiCo, with red regions in the heatmaps indicating areas of high attention.
• Figure 3: Analysis of semantic anchors' impact on survival prediction.