Skip to main content

Facial recognition solves patient identification: study

Jul. 1, 2020, 10:05 AM

by Paul Govern

Patient misidentification is an all too common cause of medical error.

In low- and middle-income countries, free, open-source facial recognition software could provide an economical solution for verifying patient identity across health care settings, according to a study by Martin Were, MD, MS, and colleagues, appearing in the International Journal of Medical Informatics.

Martin Were, MD, MS

At adult clinics in Western Kenya, arriving patients who agreed to the study sat for an average of 13 photographs taken using a standard camera linked to a computer. The images make up a training set for OpenFace, an open-source, deep neural network facial recognition system. At a second station within the clinic, the patients were again photographed and the system attempted to match these images against the stored representations.

The authors note that patients returning to these clinics often neglect to bring their identification cards. To match these arrivals to their medical records, the information system uses an algorithm based on demographic data, but according to the authors, given inconsistencies in capturing demographic information, that matching fails 30% to 50% of the time.

On first attempts, the facial recognition system correctly identified 102 out of the 103 patients enrolled in the study. One patient was misidentified on the first attempt but on a second try the error was corrected. Wearing of eyeglasses did not affect system performance.

“We were very pleased by the accuracy of this open-source system but are aware that such systems have to be implemented with great attention to patient confidentiality,” said Were, associate professor of Biomedical Informatics at Vanderbilt University Medical Center and a member of the Vanderbilt Institute for Global Health.

“For low- and middle-income countries, where patient identification may involve various additional challenges, automated facial recognition modules attached to electronic health records could provide a vast improvement at low cost. This study adds to the ongoing conversation about the costs and benefits of these systems in health care.”

Other researchers on the study included Sight Ampamya of Moi University in Kenya and John Kitayimbwa of Uganda Christian University. The study was supported by the Norwegian Program for Capacity Development in Higher Education and Research for Development.

Recent Stories from VUMC News and Communications Publications

Vanderbilt Medicine
VUMC Voice