Performance of an Open Source Facial Recognition System for Unique Patient Matching in a Resource-Limited Setting

Abstract

Background: The lack of unique patient identifiers is a challenge to patient care in developing countries. Probabilistic and deterministic matching approaches remain sub-optimal. However, affordable and scalable biometric solutions have not been rigorously evaluated in these settings. Methods: We implemented and evaluated performance of an open-source facial recognition system, OpenFace, integrated within a nationally-endorsed electronic health record system in Western Kenya. Patients were first enrolled via facial images, and later matched via the system. Accuracy of facial recognition was evaluated using Sensitivity; False Acceptance Rate (FAR); False Rejection Rate (FRR); Failure to Capture Rate (FTC) and Failure to Enroll Rate (FTE). 103 patients (mean age 37.8, 49.5% female) were enrolled. Results: The system had a sensitivity of 99.0%, FAR<1%, FRR 0.00, FTC 0.00 and FTE 0.00. Wearing spectacles did not affect performance. Conclusion: An open source facial recognition system correctly and accurately identified almost all patients during the first match.