The Accuracy of Artificial Intelligence Models in Carpal Tunnel Diagnosis: A Systematic Review and Meta-analysis.
Authors
Affiliations (5)
Affiliations (5)
- Temerty Faculty of Medicine, University of Toronto, Toronto, Ontario, Canada.
- Division of Plastic, Reconstructive & Aesthetic Surgery, Department of Surgery, University of Toronto, Toronto, ON, Canada.
- Faculty of Medicine, University of Ottawa, Ottawa, Ontario, Canada.
- Faculty of Health Sciences, Queen's University, Kingston, Ontario.
- College of Medicine, University of Sharjah, United Arab Emirates.
Abstract
Artificial intelligence (AI) has been integrated into diagnostic modalities like nerve conduction studies (NCS) and ultrasound (US) to improve their performance in detecting idiopathic median neuropathy at the carpal tunnel (IMNCT), and its signs and symptoms referred to as carpal tunnel syndrome (CTS). AI could be a useful tool for streamlining diagnosis to improve accessibility and efficiency of CTS diagnosis. This systematic review evaluated AI diagnostic accuracy for CTS. This review was registered with PROSPERO (CRD42024606291) and adhered to PRISMA guidelines. Searches were performed across Ovid MEDLINE, Ovid EMBASE, and Cochrane CENTRAL. Studies were included if they involved AI models (index test) in CTS diagnosis, where the reference standard was established by US, NCS, and/or clinical diagnosis. Results were synthesized via bivariate analysis to calculate pooled sensitivity, specificity, area under the curve (AUC), and positive and negative likelihood ratios (LRs). Twenty studies were included, with 17 assessing AI's diagnostic accuracy and four evaluating its ability to classify CTS severity. Half of the studies utilized US as the reference standard, followed by NCS (30%). The AI models demonstrated a sensitivity of 0.884 (95% CI, 0.862-0.903), specificity of 0.892 (95% CI, 0.859-0.918), AUC of 0.936, LR+ of 8.260 (95% CI, 6.220-10.800), and LR- of 0.131 (95% CI, 0.108-0.150) in diagnosing CTS. The overall certainty of the evidence was rated as moderate. AI models show promise in reliably diagnosing CTS, which can serve as a useful adjunct in the clinic to promote efficiency and workflow while standardizing diagnostic practices.