AI facial analysis offers new sickle cell insights - Children's National

Artificial intelligence (AI) continues to move pediatric medicine forward, and now it's showing that a simple smartphone photo may hold clues to sickle cell disease (SCD). Using machine learning, a team at Children's National identified subtle facial differences linked to SCD in 308 children and young adults in Kinshasa, Democratic Republic of the Congo (DRC). Half of them had confirmed SCD, and each was matched with a control participant of the same age and sex.

What this means

In the study, published in PLOS Digital Health, the team's model found 14 facial features that differed between the two groups. Six measurements around the eyes, nose and mouth did the most to tell participants with SCD apart from controls. Using those six, the model reached an accuracy of about 80% for the identification of children with SCD. Differences were most noticeable in older age groups.

Why it matters

SCD affects about 8 million people worldwide and is a major cause of childhood illness and death in sub-Saharan Africa. In places like the DRC, access to screening, confirmatory testing and specialist care is often limited. Non-invasive tools that use widely available smartphones could one day support standard clinical and community care.

Children's National leads the way

To the team's knowledge, this is the first machine learning study of facial features associated with SCD. The model is designed to be interpretable so clinicians and researchers can see exactly which facial measurements drive the results.

"Sickle cell disease is most common in the places where screening and specialist care are hardest to reach, so we asked whether a tool as widely available as a smartphone camera could help the care of children with the condition," said Marius George Linguraru, DPhil, MA, MS, director of the Division of AI Research at Children's National and senior author of the study. "Because our machine learning model is interpretable, we can see exactly which facial features are associated with sickle cell disease. That transparency is essential as we build digital tools meant to support, not replace, lab testing and clinical care."

The work builds on the team's earlier facial analysis research in populations with rare diseases that are underrepresented in digital health, including children in the DRC.

What's next for the team

The team will need studies across multiple sites, over time and with detailed clinical data for further analysis. Those studies would test whether these facial features could serve as digital biomarkers for the non-invasive monitoring of disease progression and early detection of complications.

Children's National led this research in collaboration with partners in the U.S., the Democratic Republic of Congo, Malawi and the United Kingdom, including the University of California, Irvine; Texas A&M University; Nationwide Children's Hospital; the University of Glasgow; Kamuzu University of Health Sciences; George Washington University; the University of Colorado Anschutz Medical Campus; and Oregon Health & Science University.

Additional Children's National authors for the study include D'Andre Spencer, MPH; Xinyang Liu, PhD; Matthew S Bramble, PhD; Faisal Al Munajjed; and Syed Muhammad Anwar, PhD, MS.

You can read the full study, "Quantitative Facial Phenotyping and Age-Associated Differences in Sickle Cell Disease Using Interpretable Machine Learning," in PLOS Digital Health.