Measuring What VLMs Don't Say: Validation Metrics Hide Clinical Terminology Erasure in Radiology Report Generation
This addresses a critical blind spot in evaluating radiology report generation models for clinical deployment, highlighting risks of metric gaming and demographic bias.
The paper tackled the problem of Vision-Language Models in radiology generating repetitive, generic text that omits clinical terminology despite high validation scores, and introduced Clinical Association Displacement and Weighted Association Erasure to quantify this erasure and signal loss across demographic groups.
Reliable deployment of Vision-Language Models (VLMs) in radiology requires validation metrics that go beyond surface-level text similarity to ensure clinical fidelity and demographic fairness. This paper investigates a critical blind spot in current model evaluation: the use of decoding strategies that lead to high aggregate token-overlap scores despite succumbing to template collapse, in which models generate only repetitive, safe generic text and omit clinical terminology. Unaddressed, this blind spot can lead to metric gaming, where models that perform well on benchmarks prove clinically uninformative. Instead, we advocate for lexical diversity measures to check model generations for clinical specificity. We introduce Clinical Association Displacement (CAD), a vocabulary-level framework that quantifies shifts in demographic-based word associations in generated reports. Weighted Association Erasure (WAE) aggregates these shifts to measure the clinical signal loss across demographic groups. We show that deterministic decoding produces high levels of semantic erasure, while stochastic sampling generates diverse outputs but risks introducing new bias, motivating a fundamental rethink of how "optimal" reporting is defined.