Is There Knowledge Left to Extract? Evidence of Fragility in Medically Fine-Tuned Vision-Language Models

TL;DR

This study evaluates the robustness and reasoning capabilities of medical vision-language models, revealing their fragility, prompt sensitivity, and limited benefit from domain-specific fine-tuning in high-stakes medical tasks.

Contribution

It provides a comprehensive analysis showing that medical fine-tuning does not reliably enhance reasoning and highlights the models' sensitivity to prompt variations and visual representation weaknesses.

Findings

Performance drops to near-random with increased task difficulty.

Medical fine-tuning offers no consistent performance advantage.

Models are highly sensitive to prompt formulation, affecting accuracy.

Abstract

Vision-language models (VLMs) are increasingly adapted through domain-specific fine-tuning, yet it remains unclear whether this improves reasoning beyond superficial visual cues, particularly in high-stakes domains like medicine. We evaluate four paired open-source VLMs (LLaVA vs. LLaVA-Med; Gemma vs. MedGemma) across four medical imaging tasks of increasing difficulty: brain tumor, pneumonia, skin cancer, and histopathology classification. We find that performance degrades toward near-random levels as task difficulty increases, indicating limited clinical reasoning. Medical fine-tuning provides no consistent advantage, and models are highly sensitive to prompt formulation, with minor changes causing large swings in accuracy and refusal rates. To test whether closed-form VQA suppresses latent knowledge, we introduce a description-based pipeline where models generate image descriptions that a text-only model (GPT-5.1) uses for diagnosis. This recovers a limited additional signal but remains bounded by task difficulty. Analysis of vision encoder embeddings further shows that failures stem from both weak visual representations and downstream reasoning. Overall, medical VLM performance is fragile, prompt-dependent, and not reliably improved by domain-specific fine-tuning.