Uncertainty Is Not a Safety Net for Clinical VQA, but Can It Anticipate Model Failure?
For clinicians and developers of clinical VLMs, the paper shows that current UE methods are unreliable for safety but can serve as diagnostic indicators of model fragility.
The paper evaluates 8 uncertainty estimation methods across 12 clinical VQA models and finds that UE quality degrades with model accuracy, failing to flag errors under NOTA perturbations. However, uncertainty on unperturbed inputs can predict which predictions will collapse, suggesting UE as a diagnostic tool for model fragility.
Safe deployment of clinical vision-language models (VLMs) requires reliable uncertainty estimation (UE): a signal indicating when predictions should be trusted or escalated to a clinician. We test whether current UE methods actually deliver this signal. Benchmarking 8 methods across 12 VLMs on clinical visual question-answering (VQA), we find that UE quality is not an intrinsic property of the UE method: it tracks model accuracy, degrading precisely where the model performance is weakest, and therefore where reliability is most needed. When we stress-test models by hiding the correct option among the multiple-choice answers (NOTA perturbations), accuracy collapses while uncertainty barely changes, leaving models systematically miscalibrated. Yet, we find that uncertainty on the unperturbed input reliably anticipates which predictions will collapse under NOTA, indicating that UE in current VLMs carries diagnostic information about model fragility. Our results position UE as a diagnostic tool for identifying fragile predictions and motivate perturbation-based evaluation as a path toward safe clinical deployment.