Back to 2026 Abstracts
Multicenter Prospective Cohort Study For The Evaluation Of NerveAI; An AI Algorithm-based Application For Clinical Decision Support In Patients With Headaches And Nerve Pain
Giulia Mönnink, MD1, Weishen Pan, PhD
2, Merel H J Hazewinkel, MD, PhD
1, Ahmed M. Afifi, MD
3, William G. Austen, MD
4, Thomas R. Campion, PhD
5, Robert Hagan, MD
6, Jeffrey Janis, MD
7, Rohan Jotwani, MD
8, Salam A. Kassis, MD
9, Cassandra Ligh, MD
10, Eric Liao, MD, PhD
10, Angelo A. Leto Barone, MD
11, Paul G. Matthew, MD
12, Ziv Peled, MD
13, Fei Wang, PhD
2, Lisa Gfrerer, MD, PhD
1.
1Weill Cornell Medicine, Division of Plastic and Reconstructive Surgery, New York, NY, USA,
2Weill Cornell Medicine, Department of Population Health Sciences, New York, NY, USA,
3University of Wisconsin School of Medicine and Public Health, Department of Plastic and Reconstructive Surgery, Madison, WI, USA,
4Massachusetts General Hospital, Harvard Medical School, Division of Plastic and Reconstructive Surgery, Boston, MA, USA,
5Weill Cornell Medicine, Clinical & Translational Science Center, New York, NY, USA,
6Neuropax clinic, St. Louis, MO, USA,
7The Ohio State University Wexner Medical Center, Department of Plastic and Reconstructive Surgery, Columbus, OH, USA,
8Weill Cornell Medicine, Department of Anesthesiology, New York, NY, USA,
9The Vanderbilt Clinic, Department of Plastic and Reconstructive Surgery, Nashville, TN, USA,
10Children's Hospital of Philadelphia, Department of Pediatric Plastic and Reconstructive surgery, Philadelphia, PA, USA,
11Nemours Children’s Hospital, Division of Plastic and Craniofacial Surgery, Orlando, FL, USA,
12Mass General Brigham Health, Department of Neurology, Foxborough, MA, USA,
13Peled Surgery, San Francisco, CA, USA.
PURPOSE:Diagnosis of nerve pain in patients with headache disorders (HDs) at the initial point of care (IPOC) remains challenging, often delaying surgical treatment. In a retrospective study using a training dataset, we demonstrated that NerveAI, a machine learning-based screening tool, can identify nerve pain using pattern recognition (AUROC 0.879). This current study prospectively evaluates NerveAI.
METHODS: This prospective, multi-institutional study recruited headache patients from specialized nerve clinics. Subjects reported prior headache diagnoses from IPOC providers and drew their pain on a web-based 3D head model. Each drawing was evaluated by a) a physician blinded to clinical information with expertise in neuralgia (neurologist/headache specialist or headache surgery-trained surgeon) and b) NerveAI. Subjects then underwent clinical evaluation with the treating physician, and diagnoses were made using ICHD-3 criteria. Diagnostic accuracy was compared among a) blinded experts b) NerveAI, and c) treating physicians (gold standard).
RESULTS:Among 26 patients, 15 (58%) had a prior neuralgia diagnosis from IPOC providers. After clinical evaluation, 25 (96%) were diagnosed with neuralgia. Blinded expert assessment of pain drawings achieved 100% diagnostic accuracy. NerveAI demonstrated strong region-specific accuracy: occipital (96.1%; AUROC 0.977), temporal (88.5%; AUROC 0.908), and frontal (84.6%; AUROC 0.899).
CONCLUSION: Blinded expert physicians achieve good diagnostic accuracy assessing pain drawings, substantially outperforming IPOC providers. NerveAI showed promising results for region-specific nerve pain diagnosis. Validation with a larger population is needed.
Back to 2026 Abstracts
