MEDLAYXPLAIN: Benchmarking the Expert-Lay Gap in Medical Vision-Language Models

The article introduces MedLayXPlain, a large-scale benchmark for Medical Lay Language Generation (MLLG) consisting of 122,789 samples across 8 imaging modalities. It employs a three-step pipeline called HOVER for generating lay captions from expert descriptions, ensuring semantic accuracy and reducing hallucinations. Benchmark results highlight a significant Expert-Lay Gap, indicating that while medical Vision-Language Models (VLMs) excel in expert captioning, they struggle with patient-accessible language, underscoring the need for improved models to facilitate effective patient communication and education.