Vanderbilt’s Eric Tkaczyk, MD, PhD, has received a five-year, $4.2 million grant from the National Heart, Lung, and Blood Institute, part of the National Institutes of Health (NIH), to validate an artificial intelligence technology for measuring skin changes in patients with chronic graft-versus-host disease (cGVHD).
cGVHD is a common complication following hematopoietic stem cell transplantation, a potentially curative treatment for life-threatening blood diseases. The condition often manifests as skin redness and inflammation, but current assessment methods are subjective and time-consuming.
“This grant will allow us to refine and validate an AI tool that can accurately and efficiently measure cGVHD skin changes from patient photographs. By improving the consistency, objectivity and efficiency of cGVHD assessments, we hope to enhance clinical trials and patient outcomes,” said Tkaczyk, assistant professor of Dermatology, Electrical and Computer Engineering, Biomedical Engineering and Biomedical Informatics.
Tkaczyk and collaborators will assemble a database of more than 11,000 photographs and associated clinical information from diverse patient populations at five centers: Fred Hutchinson Cancer Center, Mayo Clinic, NIH, University of Pennsylvania and VUMC. The AI’s accuracy will be compared to expert dermatologist assessments and standard in-person evaluations.
The study aims to quantify and overcome potential AI biases related to skin tone, gender, photography conditions and disease severity.
If successful, the technology could enable frequent at-home monitoring, reduce the burden on health care providers, and serve as a valuable tool for observational and therapeutic studies.
“It is exhilarating to have the opportunity to validate the AI for clinical use after several years of work by the Vanderbilt team,” Tkaczyk said. The medical image processing team is led by Benoit Dawant, PhD, Cornelius Vanderbilt chair in Engineering; other team members include Andrew McNeil, PhD, and Inga Saknite, PhD.
The research is supported by NIH grant R01HL169944.
