This whitepaper examines selected AI applications in neurology: stroke imaging and workflow, EEG and epilepsy, multimodal dementia assessment, movement analysis, multiple sclerosis follow-up and document synthesis with large language models. It is a narrative synthesis of 27 sources with a main clinical cutoff of June 15, 2026, organized by clinical task and by maturity stage, from retrospective development to routine implementation. It is non-peer-reviewed gray literature published by Arkangel AI, and the evidence it cites does not validate any Arkangel product.
Stroke imaging has the most developed operational pathway in the corpus, but performance is uneven: in a comparison on scans from 500 patients, one commercial tool detected large-vessel occlusions with a sensitivity of 0.62, against 0.65–0.96 for eight human readers, which the authors note is insufficient to rule out an occlusion. A multimodal dementia model trained on 51,269 participants reached a microaveraged AUROC of 0.94 for cognitive status and 0.96 for etiologies, without established prospective effects on care. For EEG, dementia and other domains, support for review and data integration is better established than replacement of specialist judgment, and the regional sources found document stroke-care inequalities without demonstrating neurological-AI performance in Latin America.
