D145. 3D Head Model With AI Pattern Recognition for the Identification of Nerve Pain in Patients With Headache Disorders

PURPOSE: Screening for nerve pain in patients with headache disorders (HD) requires specialized clinical knowledge, which limits access to care. The aim of this study was to develop a 3D head model that allows for AI driven pattern recognition for nerve pain in patients with HD. METHODS: We retrospectively identified 381 pain drawings completed by patients. Another 924 pain drawings were generated by the study team. The pain drawings were drawn onto an online 3D head model within a web application. Drawings were labeled as ‘nerve pain’ or ‘no nerve pain’, and affected nerve(s) were identified. The input features of the ML model were constructed with coordinates of the pain drawings. The feature was passed into classifiers and model performance was evaluated using 5-fold nested cross-validation, with AUROC as primary metric. Precision under 0.9 sensitivity, specificity under 0.9 sensitivity and sensitivity under 0.9 specificity were also reported. RESULTS: Gradient boosting models demonstrated the strongest performance in identifying nerve pain patterns. For the occipital nerves, the AUROC was 0.928 (±0.023), with a precision of 0.898 (±0.032), specificity of 0.752 (±0.081), and sensitivity of 0.742 (±0.039). Frontal and temporal nerves nerves had an AUROC of 0.899 and 0.842 (±0.032), respectively. CONCLUSION: Our ML model demonstrated good performance in identifying nerve pain patterns in HD patients. This tool will allow for improved screening for nerve pain and may reduce the health and financial burdens on patients with HD through earlier detection by a larger pool of providers.