AI for Brain Disease Diagnosis

Artificial intelligence has moved out of the neurology research lab and into the reading room. Algorithms now flag suspected large-vessel occlusions before a radiologist opens the study, measure hippocampal volume in seconds rather than hours, and surface interictal spikes buried in days of continuous EEG. This track at the Neurology Conference examines what that shift actually means at the bedside — which diagnostic tasks AI performs reliably today, which claims remain unproven, and what has to happen between a promising model and a tool a clinician can defend in a tumour board or a court.

The scope here is deliberately diagnostic rather than exploratory. Where Machine Learning in Neuroscience deals with method development and discovery science, this session concentrates on models built to answer a clinical question about a specific patient: does this person have disease X, how far has it progressed, and what happens next. Work spanning Alzheimer's Disease, Stroke, Epilepsy and neuro-oncology is welcome, provided the diagnostic endpoint is explicit and the validation strategy is described.

Where AI Sits in the Neurological Diagnostic Pathway:

Not every algorithm does the same job. Submissions are grouped by the decision the model supports:

• Triage and prioritisation — reordering a worklist so time-critical studies surface first, as with automated LVO detection on CTA
• Detection and segmentation — locating lesions, bleeds, plaques or tumours and outlining their boundaries
• Quantification — converting an image into a number a clinician can track over time: lesion load, atrophy rate, core infarct volume
• Prognosis and progression modelling — estimating conversion from mild cognitive impairment to dementia, or seizure recurrence risk
• Workflow and reporting support — structured report generation, protocol selection, quality control on incoming scans