CPathAgent: An Agent-based Foundation Model for Interpretable High-Resolution Pathology Image Analysis Mimicking Pathologists' Diagnostic Logic

TL;DR

CPathAgent addresses the interpretability gap in pathology AI by simulating pathologists' diagnostic workflow as an agent that navigates whole-slide images (WSIs) across scales. It introduces a three-stage reasoning pipeline—global screening, navigation planning, and multi-scale reasoning—together with PathMMU-HR², a expert-validated benchmark for huge-region analysis. The framework is trained in three stages to fuse patch-level perception with agent-based WSI reasoning, and evaluated across patch, huge-region, and WSI tasks, consistently outperforming both general-purpose and pathology-specific baselines while offering transparent, step-by-step rationales. The work demonstrates that agent-based diagnostic reasoning can achieve strong accuracy with improved interpretability, paving a path toward clinically usable AI-assisted pathology tools and highlighting directions for scaling data and authentic pathologist-driven paths.

Abstract

Recent advances in computational pathology have led to the emergence of numerous foundation models. These models typically rely on general-purpose encoders with multi-instance learning for whole slide image (WSI) classification or apply multimodal approaches to generate reports directly from images. However, these models cannot emulate the diagnostic approach of pathologists, who systematically examine slides at low magnification to obtain an overview before progressively zooming in on suspicious regions to formulate comprehensive diagnoses. Instead, existing models directly output final diagnoses without revealing the underlying reasoning process. To address this gap, we introduce CPathAgent, an innovative agent-based approach that mimics pathologists' diagnostic workflow by autonomously navigating across WSI based on observed visual features, thereby generating substantially more transparent and interpretable diagnostic summaries. To achieve this, we develop a multi-stage training strategy that unifies patch-level, region-level, and WSI-level capabilities within a single model, which is essential for replicating how pathologists understand and reason across diverse image scales. Additionally, we construct PathMMU-HR2, the first expert-validated benchmark for large region analysis. This represents a critical intermediate scale between patches and whole slides, reflecting a key clinical reality where pathologists typically examine several key large regions rather than entire slides at once. Extensive experiments demonstrate that CPathAgent consistently outperforms existing approaches across benchmarks at three different image scales, validating the effectiveness of our agent-based diagnostic approach and highlighting a promising direction for computational pathology.

Figures (34)

• Figure 1: Comparison of the "black box" traditional MIL approach, LMM approach, and our proposed CPathAgent for analyzing pathology images. CPathAgent interpretably mimics pathologists’ reasoning by performing operational actions (e.g., zooming in, moving the view) while describing analytical logic.
• Figure 2: Illustration of CPathAgent framework, which mimics the diagnostic workflow of pathologists via global screening, navigation planning, and multi-scale reasoning across sequential views.
• Figure 3: Overview of the generation process for CPathAgent's navigation planning subset and multi-scale multi-view sequence reasoning instruction-tuning data.
• Figure 4: OOD results for WSI (left) and region-level (right) classification tasks. Representative models from WSI and region-level approaches are selected for comparison.
• Figure A1: Overview of the VQA-oriented dataset generation process: from multi-scale description to question generation, question filtering, question-oriented navigation path generation, and VQA-related visual reasoning.