Volumetric quantification of wound healing by machine learning and optical coherence tomography in adults with type 2 diabetes: the GC-SHEALD RCT

Type 2 diabetes mellitus is associated with impaired wound healing, which contributes substantially to patient morbidity and mortality. Glucocorticoid (stress hormone) excess is also known to delay wound repair. Optical coherence tomography (OCT) is an emerging tool for monitoring healing by 'virtual biopsy', but largely requires manual analysis, which is labour-intensive and restricts data volume processing. This limits the capability of OCT in clinical research. Using OCT data from the GC-SHEALD trial, we developed a novel machine learning algorithm for automated volumetric quantification of discrete morphological elements of wound healing (by 3mm punch biopsy) in patients with type 2 diabetes. This was able to differentiate between early / late granulation tissue, neo-epidermis and clot structural features and quantify their volumetric transition between day 2 and day 7 wounds. Using OCT, we were able to visualize differences in wound re-epithelialisation and re-modelling otherwise indistinguishable by gross wound morphology between these time points. Automated quantification of maximal early granulation tissue showed a strong correlation with corresponding (manual) GC-SHEALD data. Further, % re-epithelialisation was improved in patients treated with oral AZD4017, an inhibitor of systemic glucocorticoid-activating 11{beta}-hydroxysteroid dehydrogenase type 1 enzyme action, with a similar trend in neo-epidermis volume. Through the combination of machine learning and OCT, we have developed a highly sensitive and reproducible method of automated volumetric quantification of wound healing. This novel approach could be further developed as a future clinical tool for the assessment of wound healing e.g. diabetic foot ulcers and pressure ulcers.