Reproduce “MedXplain-VQA: Multi-Component Explainable Medical Visual Question Answering” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Hai-Dang Nguyen, Minh-Anh Dang, Minh-Tan Le, Minh-Tuan Le
MedXplain-VQA introduces a multi-component framework for explainable medical visual question answering, integrating a fine-tuned BLIP-2 model with medical query reformulation, enhanced visual attention, and structured Chain-of-Thought reasoning. The system achieves a composite explainability score of 0.683 on the PathVQA dataset, representing an 80.8% improvement over the baseline, utilizing a newly established medical-domain-specific evaluation framework.
Key insights
- Systematic integration of multiple explainable AI components (query enhancement, visual attention, structured reasoning) is crucial for achieving high transparency in medical VQA.
- Medical query reformulation significantly grounds questions in a clinical context, providing a substantial initial boost to explainability and overall performance.
- Structured Chain-of-Thought reasoning, which emulates clinical diagnostic workflows, yields the highest gains in explainability and reasoning confidence for medical VQA.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.