Computers could use subtle facial features to recognize congenital adrenal hyperplasia

Reviewed by Emily Henderson, B.Sc.November 18, 2020

Congenital adrenal hyperplasia (CAH) is a disorder that affects the adrenal gland’s ability to release hormones that regulate the body’s response to stress and disease. CAH is treatable, but it can be potentially life-threatening during illness or if not managed. The disorder is difficult to identify and much more needs to be understood about the condition. But new research conducted at the Children’s Hospital of Los Angeles has shown that computers may be able to use subtle facial features to recognize CAH. This discovery could lead to better identification of the disorder and better care of patients with CAH.

In endocrinology, CAH is one of the few emergency conditions we encounter. It is the leading cause of adrenal insufficiency in children, which means the body cannot produce aldosterone, adrenaline and cortisol. “

Mimi Kim, MD, MSc, co-director of the CAH Comprehensive Care Clinic at Children’s Hospital of Los Angeles

These hormones allow the body to manage blood pressure and respond to seizures. Additionally, CAH is characterized by higher levels of testosterone, the sex hormone. This can lead to changes in the genitals for female patients. But testosterone has another effect not directly related to sex or gender, an effect that could be used to identify CAH.

“It is quite well known that hormones like testosterone help shape facial features,” says Dr. Kim. “Because CAH causes high testosterone levels during development, it is obvious that even subtle differences could be present in patients with CAH.” This, she says, led her to wonder if facial morphology – a collection of physical traits – could help doctors identify patients with CAH.

“There was still no established link between CAH and facial morphology,” says Dr. Kim. This may be because the facial differences are subtle enough to be ignored by most doctors. “But advances in machine learning have come a long way,” he says, “especially in facial recognition.”

Dr. Kim collaborated with engineers and scientists from the University of Southern California Institute of Information Sciences to design and test her hypothesis. The team uploaded images of 102 CAH patients and 144 control individuals into computers trained in facial recognition. Through machine learning, computers were able to identify subtle differences in facial morphology and correctly identify patients with CAH with greater than 90% accuracy.

The study represents an important step on the path to better identification and understanding of CAH. The findings establish, for the first time, that not only is there a link between facial morphology and CAH, but computers can detect this link and predict CAH based on patients’ facial characteristics.

Although infants are regularly screened for CAH, genetic testing is expensive and difficult to obtain, and it is not easy to characterize the severity of the condition. “We really need a more sensitive and simpler way,” says Dr. Kim. “My hope is that this is the case; that we can use the best that technology has to offer to better understand CAH and help our patients.”

The study was published in the scientific journal JAMA Network Open. The first authors of the study were Wael AbdAlmageed and Hengameh Mirzaalian of the USC Information Sciences Institute.