Medico 2025: Visual Question Answering for Gastrointestinal Imaging

Explainability is central to clinical AI: The challenge prioritizes Explainable AI (XAI) models that don't just answer questions about medical images but provide interpretable justifications aligned with medical reasoning, recognizing that clinicians need to understand why an AI system reaches its conclusions before trusting it in practice.

Large-scale medical VQA dataset enables progress: The newly created Kvasir-VQA-x1 dataset contains 159,549 question-answer pairs across 6,500 gastrointestinal endoscopy images, providing researchers with a substantial benchmark to develop and test AI systems on clinically relevant questions rather than generic image analysis tasks.

Multimodal explanations support clinical decision-making: The challenge's second subtask requires systems to generate multimodal explanations (combining text, visual indicators, and other formats) rather than single-modality answers, reflecting the reality that effective clinical communication requires multiple forms of evidence presentation.

Expert review validates AI trustworthiness: By combining quantitative performance metrics with expert-reviewed explainability assessments, the challenge acknowledges that technical accuracy alone is insufficient—medical professionals must evaluate whether AI explanations align with established clinical reasoning and best practices.

Specialized medical AI requires domain-specific evaluation: Rather than applying generic VQA benchmarks, this challenge focuses specifically on gastrointestinal imaging with clinically relevant questions, demonstrating that responsible AI development in healthcare requires task-specific datasets and evaluation criteria tailored to actual clinical workflows.