mlsanityData to 7 Oct 2026

Hidden flaws behind expert-level accuracy of multimodal GPT-4 vision in medicine

arXiv 2401.08396 ↗First posted 16 Jan 2024cs.CV cs.AI cs.CLcited by 31

Not on the map. Why?

Abstract (arXiv)

Recent studies indicate that Generative Pre-trained Transformer 4 with Vision (GPT-4V) outperforms human physicians in medical challenge tasks. However, these evaluations primarily focused on the accuracy of multi-choice questions alone. Our study extends the current scope by conducting a comprehensive analysis of GPT-4V's rationales of image comprehension, recall of medical knowledge, and step-by-step multimodal reasoning when solving New England Journal of Medicine (NEJM) Image Challenges - an imaging quiz designed to test the knowledge and diagnostic capabilities of medical professionals. Evaluation results confirmed that GPT-4V performs comparatively to human physicians regarding multi-choice accuracy (81.6% vs. 77.8%). GPT-4V also performs well in cases where physicians incorrectly answer, with over 78% accuracy. However, we discovered that GPT-4V frequently presents flawed rationales in cases where it makes the correct final choices (35.5%), most prominent in image comprehension (27.2%). Regardless of GPT-4V's high accuracy in multi-choice questions, our findings emphasize the necessity for further in-depth evaluations of its rationales before integrating such multimodal AI models into clinical workflows.

Key facts. Cited by 31 papers in the map (data through 7 Oct 2026). Builds on 11 of its 17 references that are in the map. First posted on arXiv 16 Jan 2024.

Builds on

11 of 17 references.

References · 11

  1. 2023GPT-4 Technical Reportcited by 18,132
  2. 2022Large Language Models Encode Clinical Knowledgecited by 1,807
  3. 2023Capabilities of GPT-4 on Medical Challenge Problemscited by 630
  4. 2023Can Generalist Foundation Models Outcompete Special-Purpose Tuning? Case Study in Medicinecited by 319
  5. 2022Can large language models reason about medical questions?cited by 194
  6. 2023Opportunities and Challenges for ChatGPT and Large Language Models in Biomedicine and Healthcited by 118
  7. 2023Can GPT-4V(ision) Serve Medical Applications? Case Studies on GPT-4V for Multimodal Medical Diagnosiscited by 77
  8. 2023Matching Patients to Clinical Trials with Large Language Modelscited by 68
  9. 2023Multimodal ChatGPT for Medical Applications: an Experimental Study of GPT-4Vcited by 38
  10. 2023Multimodal Foundation Models Exploit Text to Make Medical Image Predictionscited by 19
  11. 2023PubMed and Beyond: Biomedical Literature Search in the Age of Artificial Intelligencecited by 16

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2024: 9; 2025: 14; 2026: 8.

31 papers.

Sort:

By area:

  1. 2024Benchmarking Large Language Models on Answering and Explaining Challenging Medical Questionscited by 106
  2. 2025MedCaseReasoning: Evaluating and learning diagnostic reasoning from clinical case reportscited by 34
  3. 2024ClinicalBench: Can LLMs Beat Traditional ML Models in Clinical Prediction?cited by 31
  4. 2024Superhuman performance of a large language model on the reasoning tasks of a physiciancited by 28
  5. 2025Automating Expert-Level Medical Reasoning Evaluation of Large Language Modelscited by 21
  6. 2024Pattern Recognition or Medical Knowledge? The Problem with Multiple-Choice Questions in Medicinecited by 13
  7. 2025Your other Left! Vision-Language Models Fail to Identify Relative Positions in Medical Imagescited by 12
  8. 2025From Scores to Steps: Diagnosing and Improving LLM Performance in Evidence-Based Medical Calculationscited by 11
  9. 2025The Illusion of Readiness in Health AIcited by 10
  10. 2024A Perspective for Adapting Generalist AI to Specialized Medical AI Applications and Their Challengescited by 8
  11. 2025Capabilities of GPT-5 across critical domains: Is it the next breakthrough?cited by 6
  12. 2024Process-Supervised Reward Models for Verifying Clinical Note Generation: A Scalable Approach Guided by Domain Expertisecited by 5
19 more
  1. 2024MultiViewDx: Evidence-Linked Multi-View Clinical Diagnosiscited by 4
  2. 2025Uncertainty-Driven Expert Control: Enhancing the Reliability of Medical Vision-Language Modelscited by 4
  3. 2024Entry-level guide to the use of large language models for medical researchcited by 3
  4. 2025GRAID: Enhancing Spatial Reasoning of VLMs Through High-Fidelity Data Generationcited by 3
  5. 2026Medical thinking with multiple imagescited by 3
  6. 2025A Real-World Evaluation of LLM Medication Safety Reviews in NHS Primary Carecited by 2
  7. 2026CT-Bench: A Benchmark for Multimodal Lesion Understanding in Computed Tomographycited by 2
  8. 2025SynthEHR-Eviction: Enhancing Eviction SDoH Detection with LLM-Augmented Synthetic EHR Datacited by 1
  9. 2025Beyond Diagnosis: Evaluating Multimodal LLMs for Pathology Localization in Chest Radiographscited by 1
  10. 2026MedArena: Comparing LLMs for Medicine-in-the-Wild Clinician Preferencescited by 1
  11. 2024Constructive Apraxia: An Unexpected Limit of Instructible Vision-Language Models and Analog for Human Cognitive Disorderscited by 0
  12. 2025PCS: Perceived Confidence Scoring of Black Box LLMs with Metamorphic Relationscited by 0
  13. 2025Vision Language Models versus Machine Learning Models Performance on Polyp Detection and Classification in Colonoscopy Imagescited by 0
  14. 2025CheXPO-v2: Preference Optimization for Chest X-ray VLMs with Knowledge Graph Consistencycited by 0
  15. 2026SleepVLM: A Rule-Grounded Vision-Language Model for Auditable Sleep Stagingcited by 0
  16. 2026Captioning Daily Activity Images in Early Childhood Education: Benchmark and Algorithmcited by 0
  17. 2026Agentic systems for breast cancer treatment recommendationscited by 0
  18. 2026CardioBench: A Real-World Data Benchmark for Evaluating Large Language Models in Clinically Authentic Cardiovascular Care Scenarioscited by 0
  19. 2026Policy Iteration with Human Feedback: Bringing Post-Training RL to In-context Learningcited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2023Enhancing Medical Task Performance in GPT-4V: A Comprehensive Study on Prompt Engineering Strategies
  2. 2026Evaluating GPT-5 as a Multimodal Clinical Reasoner: A Landscape Commentary
  3. 2024GPT-4V Cannot Generate Radiology Reports Yet
  4. 2023Holistic Evaluation of GPT-4V for Biomedical Imaging
  5. 2025Capabilities of GPT-5 on Multimodal Medical Reasoning
  6. 2026MedQA-MM: Shortcuts Behind Medical Visual Reasoning
  7. 2025Radiology's Last Exam (RadLE): Benchmarking Frontier Multimodal AI Against Human Experts and a Taxonomy of Visual Reasoning Errors in Radiology
  8. 2023GPT-4V(ision) as a Generalist Evaluator for Vision-Language Tasks