🤖 AI Summary
This work proposes the first systematic framework to enhance the interpretability of Transformer-based deep learning models in computational pathology by integrating class visualizations (CVs) and activation atlases (AAs) to analyze morphological features learned across multi-granular tissue and cancer classification tasks. The approach is rigorously validated through expert pathologist assessments and quantitative metrics, including Fleiss’ Kappa for inter-rater agreement, as well as perceptual and distributional similarity measures. Findings reveal that CVs remain interpretable in tissues with pronounced morphological distinctions, while AAs uncover a hierarchical, dependency-aware representational structure. Critically, strong correlations between expert consensus and atlas separability demonstrate that model representations effectively capture the inherent complexity of pathological patterns, thereby establishing a novel expert-in-the-loop paradigm for evaluating interpretability in medical AI.
📝 Abstract
The rapid adoption of transformer-based models in computational pathology has enabled prediction of molecular and clinical biomarkers from H&E whole-slide images, yet interpretability has not kept pace with model complexity. While attribution- and generative-based methods are common, feature visualization approaches such as class visualizations (CVs) and activation atlases (AAs) have not been systematically evaluated for these models. We developed a visualization framework and assessed CVs and AAs for a transformer-based foundation model across tissue and multi-organ cancer classification tasks with increasing label granularity. Four pathologists annotated real and generated images to quantify inter-observer agreement, complemented by attribution and similarity metrics. CVs preserved recognizability for morphologically distinct tissues but showed reduced separability for overlapping cancer subclasses. In tissue classification, agreement decreased from Fleiss k = 0.75 (scans) to k = 0.31 (CVs), with similar trends in cancer subclass tasks. AAs revealed layer-dependent organization: coarse tissue-level concepts formed coherent regions, whereas finer subclasses exhibited dispersion and overlap. Agreement was moderate for tissue classification (k = 0.58), high for coarse cancer groupings (k = 0.82), and low at subclass level (k = 0.11). Atlas separability closely tracked expert agreement on real images, indicating that representational ambiguity reflects intrinsic pathological complexity. Attribution-based metrics approximated expert variability in low-complexity settings, whereas perceptual and distributional metrics showed limited alignment. Overall, concept-level feature visualization reveals structured morphological manifolds in transformer-based pathology models and provides a framework for expert-centered interrogation of learned representations across label granularities.