NeuroMCP-Agent reads EEG and flags uncertainty in its diagnoses

Researcher Rakesh Kumar Reddy Ambati of Dartmouth Health has built NeuroMCP-Agent, an AI system that screens EEG recordings for seven neurological and mental-health conditions while estimating confidence and explaining its decisions. Results come from a single centre and still require independent, multi-centre validation.

• Accuracy reached 100% for Parkinson's in the main evaluation and 92.4% under leave-one-out testing
• The system stacks 15 algorithms over 47 EEG features using 4-second windows with 75% overlap
• Reported governance score against EU AI Act, FDA and HIPAA expectations was 0.91 out of 1
• Removing data augmentation cut accuracy by 3.21 points; a single algorithm instead of the ensemble cost 5.77 points