MM-NeuroOnco: A Multimodal Benchmark and Instruction Dataset for MRI-Based Brain Tumor Diagnosis

Researchers introduce MM-NeuroOnco, a large-scale multimodal dataset containing 24,726 MRI slices and 200,000 instructions for training AI models in brain tumor diagnosis. The benchmark reveals significant challenges in medical AI, with even advanced models like Gemini 3 Flash achieving only 41.88% accuracy on diagnostic questions.

• →MM-NeuroOnco dataset includes 24,726 MRI slices from 20 sources with semantically enriched multimodal instructions for brain tumor diagnosis.
• →Current AI models struggle with medical diagnosis, as Gemini 3 Flash achieved only 41.88% accuracy on brain tumor diagnostic questions.
• →The researchers developed NeuroOnco-GPT which showed 27% absolute accuracy improvement after fine-tuning on the dataset.
• →A multi-model collaborative pipeline was created to automate medical information completion and quality control for dataset generation.
• →The benchmark includes a rejection-aware evaluation setting to reduce biases from closed-ended question formats.