Purpose

Patient selection is crucial for optimizing outcomes in headache surgery. Pain drawings are an important tool for patient screening, but interpretation requires extensive training. We sought to create a machine learning (ML) model, integrated into a mobile application, to facilitate screening for providers.

Methods

115 pain drawings created by patients were analyzed and categorized for presence of nerve pain, anatomic distribution of pain, as well as candidacy for surgery by two headache surgery experts (WGA and LG). A random forest ML algorithm was developed based on this analysis and trained using 3208 copies of the initial patient drawings. The algorithm was asked to evaluate 200 randomly selected drawings. Android Studio 2.0 was used to create a mobile application.

Results

The algorithm correctly determined nerve pain, the anatomic pain distribution of pain, and candidacy for surgery in 97.5% of test drawings (195/200). Across the different anatomic pain distributions, the mean ± SD accuracy was 98.4 ± 1.3%. The mobile application surface enabled non-surgical providers to successfully use the algorithm.

Conclusion

ML can be used in the interpretation of patient pain drawings. The integration of this screening tool into a mobile application will enable non-surgical providers such as neurologists and primary care physicians, as well patients to screen for nerve related headache increasing the likelihood of timely referral to a qualified headache surgeon.