Background  
While Diabetic Foot Ulcer (DFU) prevention and treatment have improved dramatically, assessment of DFU healing remains outdated, presenting a problem for research and clinical practice. Recent new mobile applications for standardizing DFU images have been reported, however, they are not automated and require large manual efforts.

Aim  
This study aims to define a reliable and sensitive metric by using machine learning to quantify wound severity, taking into account its size, shape, ischaemic and infection severity using a standard 2D photograph. This single metric could be used to identify and triage non-healing DFUs, as well as track wound healing in response to treatment.

Method  
Serial photographs of DFUs in 28 patients (90% T2DM, mean age 63, 83% male) with varying severity of DFU (72% UT 1A-C), were collected and analysed. Images from sequential clinic visits were automatically aligned. Wounds were automatically localized and segmented.  
  


Results  
We have characterized DFUs and quantified changes in DFU characteristics across visits. 91% of baseline images were correctly identified and localized by the artificial intelligence (AI)-based wound localisation method.

Discussion  
If successful, this measure will facilitate targeting of appropriate therapies to individual patients, improving healing outcomes and therefore resulting in fewer amputations.



DFU Track: Building a novel, Artificial Intelligence-based method to monitor wound healing.