A novel evaluation framework inspired by medical cognitive tests assesses how well AI systems understand and reason about complex visual scenes.

• Evaluates eight distinct reasoning capabilities in Large Vision-Language Models
• Creates a benchmark of 251 richly annotated images for comprehensive testing
• Draws from established medical cognitive assessment techniques (Cookie Theft task)
• Provides structured measurement of AI's ability to process visual information cognitively

This framework has significant medical applications, potentially transforming how we evaluate AI systems for clinical decision support, cognitive assessment automation, and patient monitoring technologies.

A Cognitive Evaluation Benchmark of Image Reasoning and Description for Large Vision-Language Models