AI-driven OCT imaging system enables accurate wound healing assessment

Regardless of size or severity, human skin wounds are difficult to observe until they heal. Because biopsies destroy the wound site, they are too invasive to be repeatedly monitored on a regular basis. Additionally, most of the medical imaging equipment that performs that role is large, expensive, and often reserved for more urgent diagnoses. Clinicians typically rely on visual inspection or rapid measurement of wound size over time.

Biomedical engineers at Duke University are developing a solution based on research completed as part of a multi-year collaboration with Nokia Bell Laboratories. Researchers have shown that by using a custom-built optical coherence tomography (OCT) imaging system in conjunction with an artificial intelligence (AI) model based on a deep understanding of tissue regeneration, it is possible to accurately and objectively measure the progress of wound healing over time.

Using a new approach, the researchers also show that the hydrogels they are developing to improve wound healing work more effectively with stiffer mechanical properties. This result has a two-fold benefit in a difficult field for both clinicians and researchers.

The study will be published online on March 20 in the journal Cellular biomaterials.

Wound healing is a complex process, and what you see on the surface does not always reflect what is happening underneath. For more than a decade, my lab has been developing hydrogel-based treatments that guide tissue healing and regeneration. Our partnership with Nokia Bell Labs allows us to combine advanced optical imaging with AI to gain unprecedented insight into how biomaterials induce healing beneath the surface. ”

Sharon Gerecht, Duke University Professor of Biomedical Engineering and Professor Paul M. Gross

OCT is best known for its role in eye care, providing 3D images of the back of the eye to help diagnose and monitor retinal diseases. Now, researchers are applying the same depth-resolved imaging capabilities to wound healing, using light to non-invasively visualize tissue structures and blood flow beneath the skin.

However, imaging alone is not enough to turn these rich scans into meaningful biological insights. Analyzing information requires quantitative tools that can quickly interpret large amounts of complex data. A collaboration with Nokia Bell Labs proved essential.

Through a multi-year project, Nokia Bell Labs researchers developed a custom OCT system with AI-driven analysis methods trained on image datasets acquired at the Gerecht Institute. This OCT-AI platform enabled the team to go beyond simple visualization to automatically quantify how tissue architecture and vascular dynamics evolve over time and objectively assess the extent of healing.

To evaluate the technology, the collaborative team applied it to wounds in mice treated with a hydrogel developed in Gerecht’s lab. To demonstrate the broader research potential of this platform, they compared hydrogels with relatively soft and relatively stiff mechanical properties.

Over the course of two weeks, this platform allowed a detailed internal look at how granulation tissue, the smooth glassy tissue that initially fills the wound, fills the space and matures. The data showed that stiffer hydrogels helped form more initial granulation tissue in a shorter time and also helped the initial tissue transition into intact regenerating tissue faster.

“With the technology we developed, we were able to monitor blood flow near a wound and collectively understand the structural and vascular changes occurring in real time,” said Jiyoung Sung, a postdoctoral researcher in Gerecht’s lab and co-lead author of the paper. “Thanks to AI, we were able to quantitatively track those changes and get more objective results, rather than trying to manually analyze images.”

Going forward, the research collaboration plans to continue developing this platform for potential clinical use. Although the OCT-AI platform has proven effective in this relatively simple scenario, more work is needed to enable it to not only monitor healing progress but also predict various disease states. For example, Gerecht and her team plan to pursue funding for research aimed at building the system to predict the healing of chronic wounds in patients with diabetes.

This research was supported by the P30 Cancer Center Support Grant (P30 CA014236), the American Heart Association, Duke Regenerative Center (DRC), Duke Science and Technology (DST), and Nokia Bell Laboratories.