AI Is Analyzing Faces to Aid Rare Disease Diagnosis

Imagine facing this: You’ve been ill for months, and no doctor can determine the cause. Countless hospital visits and exams do not produce any tangible results. The very idea seems ludicrous to many, but for the few living with a rare disease, it’s all too real.

According to the Genetic and Rare Disease Information Center, there might be as many as 7000 rare diseases, with the total number of Americans living with a rare disease estimated to be as high as 30 million.

But a technique using computer-aided facial analysis could lend a hand in diagnosing at least 1 rare disease. A group of researchers at the University Hospital Bonn and the Charité Universitätsmedizin in Berlin, Germany, used artificial intelligence (AI) image analysis to identify patients with GPI anchor deficiencies and presented their findings in a paper published this month in the journal Genome Medicine.

Using data on genetic material of cells, surface texture, and typical facial features, researchers employed artificial intelligence methods to simulate disease models of Mabry syndrome, a condition featuring intellectual disability. The technology could serve for diagnosis in other diseases as well, researchers said.

“The artificial modeling of gene-typical faces that we achieved with these data sets clearly shows that the computer-aided evaluation of patients' portraits can facilitate and improve the diagnosis of GPI anchor deficiencies, which is significant progress,” wrote lead author Alexej Knaus, PhD, from the Institute for Genome Statistics and Bioinformatics of the University Hospital Bonn.

GPI, or glycolipids, maintain the stability of a cell and aid in cell-to-cell communication. In GPI anchor deficiencies, the interaction is faulty because of gene mutation.

As those with Mabry syndrome have a range of distinctive facial features, including a narrow, tent-shaped upper lip, a broad nose bridge, and wide-set eyes, Peter Krawitz, MD, a co-author of the GPI study, said patients often suffer many years in uncertainty before receiving a correct diagnosis.

“Face2Gene Clinic will shorten the diagnostic odyssey for many families,” Krawitz said, referencing the name of the app that uses this technology.

In their research, the group photographed the faces of 91 patients with the condition. Krawitz said the main challenge in genomics is making sense out of the thousands of variants detected in a patient and that AI is particularly useful in interpreting that data. He believes the technology can be applicable to all conditions for which diagnoses can be done through a patient photo.

“We are quite optimistic that with AI it is possible to group patients of unknown cause that have a common mutation and the same gene,” he said.

The research group is conducting a large study on the performance that can be gained by using AI in exome analysis in collaboration with FDNA, an AI precision medicine company.

“Until recently I was the greatest skeptic about this [AI] technology, but the systems have an astonishingly good performance,” Krawitz said. “The challenge now from a scientific point of view is to understand why they are better in certain tests than humans are. So there’s something we can learn, too.”