TaxoMIL: Taxonomy-Constrained Learning for Hierarchical Whole Slide Image Analysis

2026-06-29

AI Summary

This work addresses the limitation of existing whole-slide image (WSI) diagnostic methods that treat diagnosis as a flat classification task, thereby ignoring the hierarchical structure inherent in clinical decision-making and often yielding semantically inconsistent or hierarchically conflicting predictions. To overcome this, the authors propose TaxoMIL, a novel framework that, for the first time, integrates medical taxonomic constraints into both multiple instance learning and diagnostic text generation. TaxoMIL employs a dual-head Transformer decoder to produce multi-granular diagnostic narratives, while leveraging a taxonomy-guided embedding space and a vision–language alignment mechanism to ensure clinically coherent, hierarchy-aware outputs. Evaluated on three diverse WSI datasets, TaxoMIL significantly outperforms current MIL classifiers and vision–language generation approaches, delivering diagnostically accurate and structurally consistent results.

Abstract

Whole slide image (WSI) analysis is central to computational pathology, with multiple instance learning (MIL) emerging as the standard pipeline for slide-level diagnosis. However, conventional approaches formulate WSI diagnosis as a flat classification task over discrete labels, contradicting the inherently hierarchical, coarse-to-fine nature of clinical reasoning. Although recent hierarchical classifiers and vision-language models (VLMs) have sought to address this structural gap, they either fail to capture semantic continuity between related diagnoses or suffer from unconstrained text generation that produces taxonomic hallucinations and parent-child label violations. To address these limitations, we propose TaxoMIL, a taxonomy-constrained framework that reformulates WSI diagnosis as a multi-granularity text generation task. TaxoMIL utilizes a dual-head Transformer decoder to generate coarse- and fine-level diagnostic text, and introduces taxonomy-guided objectives that explicitly structure the label embedding space and strictly ground slide-level visual representations within the clinical taxonomy. Extensive experiments across three diverse WSI datasets demonstrate that TaxoMIL consistently outperforms state-of-the-art MIL classifiers and VLM-based generative methods, yielding accurate and hierarchy-aware diagnostic predictions. The code is released at https://github.com/QuIIL/TaxoMIL

Problem

• whole slide image
• hierarchical diagnosis
• taxonomy constraint
• multiple instance learning
• computational pathology

Innovation

• Taxonomy-Constrained Learning
• Hierarchical Diagnosis
• Multiple Instance Learning
• Vision-Language Model
• Whole Slide Image Analysis