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

• URL: http://arxiv.org/abs/2604.09841v1
• Date: Fri, 10 Apr 2026 19:14:03 GMT
• Title: Is There Knowledge Left to Extract? Evidence of Fragility in Medically Fine-Tuned Vision-Language Models
• Abstract summary: We evaluate four paired open-source vision-language models (VLMs) across four medical imaging tasks.<n>We find that performance degrades toward near-random levels as task difficulty increases, indicating limited clinical reasoning.<n>Medical VLM performance is fragile, prompt-dependent, and not reliably improved by domain-specific fine-tuning.
• Score: 23.48481783960546
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• 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.

Related papers

• DeepTumorVQA: A Hierarchical 3D CT Benchmark for Stage-Wise Evaluation of Medical VLMs and Tool-Augmented Agents [15.131996289027926]
  We propose a hierarchical benchmark that decomposes 3D CT reasoning into four stages: recognition, measurement, visual reasoning, and medical reasoning.<n>The benchmark contains 476K questions across 42 clinical subtypes on 9,262 3D CT volumes.
  arXiv  Detail & Related papers  (2026-05-10T17:57:57Z)
• MedLVR: Latent Visual Reasoning for Reliable Medical Visual Question Answering [8.481971263315854]
  textscMedLVR introduces an explicit visual evidence state into autoretext decoding.<n>Latent visual reasoning provides an effective mechanism for preserving diagnostically relevant visual evidence.
  arXiv  Detail & Related papers  (2026-04-10T16:03:03Z)
• MedObvious: Exposing the Medical Moravec's Paradox in VLMs via Clinical Triage [20.835664121303534]
  Vision Language Models (VLMs) are increasingly used for tasks like medical report generation and visual question answering.<n>In clinical practice, interpretation begins with pre-diagnostic sanity checks.<n>Existing benchmarks largely assume this step is solved, and therefore miss a critical failure mode.<n>We introduce MedObvious, a 1,880-task benchmark that isolates input validation as a set-level consistency capability.
  arXiv  Detail & Related papers  (2026-03-24T17:59:54Z)
• Unleashing Video Language Models for Fine-grained HRCT Report Generation [15.816436105926542]
  AbSteering is an abnormality-centric framework that steers VideoLMs toward precise HRCT report generation.<n>Our results demonstrate that general-purpose VideoLMs possess strong transferability to high-volume medical imaging.
  arXiv  Detail & Related papers  (2026-03-12T21:34:38Z)
• Deep Expert Injection for Anchoring Retinal VLMs with Domain-Specific Knowledge [66.67024684187915]
  Large Vision Language Models (LVLMs) show immense potential for automated ophthalmic diagnosis.<n>Their clinical deployment is severely hindered by lacking domain-specific knowledge.<n>EyExIn is a framework designed to anchor retinal VLMs with expert knowledge via a Deep Expert Injection mechanism.
  arXiv  Detail & Related papers  (2026-03-07T09:43:49Z)
• Route, Retrieve, Reflect, Repair: Self-Improving Agentic Framework for Visual Detection and Linguistic Reasoning in Medical Imaging [3.6136448489318695]
  We propose R4, an agentic framework that decomposes medical imaging into four coordinated agents.<n>R4 consistently boosts LLM-as-a-Judge scores by roughly +1.7-+2.5 points and mAP50 by +2.5-+3.5 absolute points over strong single-VLM baselines.
  arXiv  Detail & Related papers  (2026-01-13T03:44:06Z)
• Med-VCD: Mitigating Hallucination for Medical Large Vision Language Models through Visual Contrastive Decoding [0.8431877864777443]
  We introduce Med-VCD, a sparse visual-contrastive decoding method that mitigates hallucinations in medical LVLMs without the time overhead of secondary decoding.<n>We show that Med-VCD raises factual accuracy by an average of 13% and improves hallucination accuracy by 6% relative to baseline medical LVLMs.
  arXiv  Detail & Related papers  (2025-12-01T17:40:03Z)
• TemMed-Bench: Evaluating Temporal Medical Image Reasoning in Vision-Language Models [54.48710348910535]
  Existing medical reasoning benchmarks primarily focus on analyzing a patient's condition based on an image from a single visit.<n>We introduce TemMed-Bench, the first benchmark designed for analyzing changes in patients' conditions between different clinical visits.
  arXiv  Detail & Related papers  (2025-09-29T17:51:26Z)
• Seeing the Trees for the Forest: Rethinking Weakly-Supervised Medical Visual Grounding [45.248265283405004]
  Current models struggle to associate textual descriptions with disease regions due to inefficient attention mechanisms and a lack of fine-grained token representations.<n>We introduce Disease-Aware Prompting (DAP), which uses the explainability map of a VLM to identify the appropriate image features.<n>DAP improves visual grounding accuracy by 20.74% compared to state-of-the-art methods across three major chest X-ray datasets.
  arXiv  Detail & Related papers  (2025-05-21T05:16:45Z)
• Structured Outputs Enable General-Purpose LLMs to be Medical Experts [50.02627258858336]
  Large language models (LLMs) often struggle with open-ended medical questions.<n>We propose a novel approach utilizing structured medical reasoning.<n>Our approach achieves the highest Factuality Score of 85.8, surpassing fine-tuned models.
  arXiv  Detail & Related papers  (2025-03-05T05:24:55Z)
• MMed-RAG: Versatile Multimodal RAG System for Medical Vision Language Models [49.765466293296186]
  Recent progress in Medical Large Vision-Language Models (Med-LVLMs) has opened up new possibilities for interactive diagnostic tools.<n>Med-LVLMs often suffer from factual hallucination, which can lead to incorrect diagnoses.<n>We propose a versatile multimodal RAG system, MMed-RAG, designed to enhance the factuality of Med-LVLMs.
  arXiv  Detail & Related papers  (2024-10-16T23:03:27Z)
• Adapting Visual-Language Models for Generalizable Anomaly Detection in Medical Images [68.42215385041114]
  This paper introduces a novel lightweight multi-level adaptation and comparison framework to repurpose the CLIP model for medical anomaly detection. Our approach integrates multiple residual adapters into the pre-trained visual encoder, enabling a stepwise enhancement of visual features across different levels. Our experiments on medical anomaly detection benchmarks demonstrate that our method significantly surpasses current state-of-the-art models.
  arXiv  Detail & Related papers  (2024-03-19T09:28:19Z)
• Robust and Interpretable Medical Image Classifiers via Concept Bottleneck Models [49.95603725998561]
  We propose a new paradigm to build robust and interpretable medical image classifiers with natural language concepts. Specifically, we first query clinical concepts from GPT-4, then transform latent image features into explicit concepts with a vision-language model.
  arXiv  Detail & Related papers  (2023-10-04T21:57:09Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.