Machine learning assisted thermal reperfusion mapping reveals real-time changes in foot microcirculation during endovascular revascularisation

Objective To test the feasibility of a novel thermal perfusion imaging and mapping system to show microcirculatory changes of plantar foot angiosomes during endovascular procedures. Methods A prospective, exploratory MedTech development study was designed to test the system in a series of patients with chronic limb-threatening ischaemia (CLTI). Intraprocedural video thermography was performed using a 320 × 240 pixel resolution, thermal camera. Initial analysis was manual, re- positioning regions of interest (ROIs) within each of five plantar foot angiosomes. Subsequent automation using deep learning segmentation allowed ROI tracking from initial frame to procedure end. Isotherms (thermal contours) created and applied to the segmented surface provides quantitative, real-time changes in the percentage distribution of isotherms overlying both treated and contralateral foot angiosomes during the procedure. Results Of 23 participants recruited, video thermography was obtained in 21 patients admitted with moderate-severe disease complexity; femoral-popliteal segment only ( n = 13), femoral-popliteal and infrapopliteal disease ( n = 5) and infrapopliteal disease only ( n = 3). Changes in percentage distribution of each isotherm overlay revealed two broad groups; plantar foot thermal perfusion responders ( n = 12) and non-responders ( n = 9). In responders, revascularisation led to a series of ‘dynamic’ phases where temperature increased in one, or more, angiosomes; ‘hottest’ isotherms expanded significantly ( p = .002) in response to endovascular procedures. In thermal perfusion non-responders, no significant change in isotherm percentage distribution was observed. Conclusion Thermal perfusion imaging is feasible using the novel implemented deep learning system. Clinicians can gain insight about microcirculatory changes at foot angiosomes in response to treatment based on three parameters; (i) temporal temperature profiles of each angiosome, (ii) real-time changes in the percentage distribution of isotherms across the plantar surface during the intervention, (iii) angiosome-specific cutaneous perfusion responses to balloon inflation and/or stent deployment. For patients with diabetes, locating the site of improved skin reperfusion has potential to predict treatment success for healing outcomes at the ulcer/woundosome