Measuring What VLMs Don't Say: Validation Metrics Hide Clinical Terminology Erasure in Radiology Report Generation

TL;DR

This paper highlights the limitations of current validation metrics for Vision-Language Models in radiology, revealing how they can mask clinical terminology omission and bias, and proposes new lexical diversity and association measures for better evaluation.

Contribution

It introduces Clinical Association Displacement (CAD) and Weighted Association Erasure (WAE), novel metrics to assess clinical fidelity and demographic fairness in radiology report generation.

Findings

Deterministic decoding leads to clinical terminology erasure.

Stochastic sampling increases diversity but may introduce bias.

Current metrics can be gamed despite high token-overlap scores.

Abstract

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.