14  Neurology and Neuropsychiatry

TipLearning Objectives

Neurological disorders present unique challenges for AI: complex symptomatology, reliance on clinical examination, and limited biomarkers for many conditions. This chapter examines AI applications across neurology and neuropsychiatry. You will learn to:

  • Evaluate AI systems for stroke detection and triage
  • Understand AI applications in neuroimaging (brain MRI, CT)
  • Assess AI tools for seizure detection and epilepsy management
  • Navigate AI-assisted neurodegenerative disease diagnosis
  • Identify AI applications in psychiatric diagnosis and treatment
  • Recognize limitations specific to neurological AI
  • Apply evidence-based frameworks for neurology AI adoption

Essential for neurologists, neurosurgeons, psychiatrists, and neuroscience care teams.

The Clinical Context: Neurology combines objective data (neuroimaging, EEG, EMG) with subjective clinical assessment (mental status, motor exam, cognitive testing). AI shows promise for pattern recognition in imaging and physiologic signals but struggles with the nuanced clinical judgment central to neurological diagnosis.

Key Applications:

14.0.1 1. Stroke AI

✅ Large Vessel Occlusion (LVO) Detection: - Multiple FDA-cleared systems (Viz.ai, RapidAI, Brainomix) - Reduces door-to-groin time by 30-50 minutes (mclell?) - Improves functional outcomes in RCTs - Published in Stroke (McLellan et al. 2022)

✅ Verdict: Strong evidence, widely deployed, clear clinical benefit.

✅ ICH Detection and Triage: - Automated intracranial hemorrhage detection - >95% sensitivity for most hemorrhage types - Prioritizes worklist, alerts neurosurgery - Published extensively (Arbabshirani et al. 2018)

✅ Verdict: FDA-cleared, clinically beneficial, widely adopted in EDs and trauma centers.

⚠️ Stroke Risk Prediction: - ML models for recurrent stroke prediction - Modest improvement over clinical scores (AUC 0.70-0.75) - Unclear how predictions change management (kamel2020machine?)

14.0.2 2. Neuroimaging AI

✅ Brain Tumor Segmentation: - Automated tumor delineation for radiation planning - Reduces planning time, improves consistency - FDA-cleared systems available (yogananda2020automated?)

⚠️ Multiple Sclerosis Lesion Detection: - Automated MS lesion counting and volumetry - Tracks disease progression - Variable performance across scanners (commowick2018objective?)

⚠️ Alzheimer’s Disease Neuroimaging: - Hippocampal volumetry, amyloid PET quantification - Predicts conversion from MCI to dementia - Not yet proven to improve clinical outcomes (park2020machine?)

14.0.3 3. Epilepsy and Seizure Detection

✅ Automated Seizure Detection: - Long-term EEG monitoring with AI-assisted review - Reduces neurologist review time - Sensitivity 85-95% for most seizure types (golmohammadi2020automatic?)

⚠️ Wearable Seizure Detectors: - Smartwatch-based seizure detection - High false positive rates - Useful for high-risk patients (SUDEP prevention) (poh2012wearable?)

14.0.4 4. Neurodegenerative Disease AI

⚠️ Parkinson’s Disease: - Smartphone-based motor assessment - Voice analysis for early detection - Modest accuracy, not diagnostic (arora2015detecting?)

⚠️ ALS Progression Prediction: - ML models predict functional decline - May inform clinical trial design - Individual-level prediction still imperfect (kueffner2015stratification?)

14.0.5 5. Psychiatric AI

❌ Depression and Suicide Risk: - Covered in Chapter 7 (Pediatrics) - Insufficient evidence for clinical deployment - Ethical concerns unresolved

⚠️ Psychosis Risk Prediction: - ML models predict conversion to psychosis in at-risk youth - Research stage only (koutsouleris2018prediction?)

Clinical Bottom Line: Neurology AI excels at acute, time-sensitive imaging interpretation (stroke, ICH) but remains limited for complex diagnosis and prognostication. Psychiatric AI faces enormous challenges—heterogeneous presentations, lack of biomarkers, stigma, and ethics. Neurologists should embrace imaging AI while maintaining skepticism about diagnostic and prognostic algorithms.

14.1 References