An exploratory, intra- and interindividual comparison of the deep neural network automatically measured calf surface radiation temperature during cardiopulmonary running and cycling exercise testing: A preliminary study

Abstract

Non-invasive and contactless infrared thermography (IRT) measurements have been claimed to indicate acute neural, cardiovascular, and thermoregulatory adaptations during exercise. Due to challenging comparability, reproducibility, and objectivity, investigations considering different exercise types and intensities, and automatic ROI analysis are currently needed. Thus, we aimed to examine surface radiation temperature (T_sr) variations during different exercise types and intensities in the same individuals, ROI, and environmental conditions. Ten healthy, active males performed a cardiopulmonary exercise test on a treadmill in the first week and on a cycling ergometer the following week. Respiration, heart rate, lactate, rated perceived exertion, the mean, minimum, and maximum T_sr of the right calf (CT_sr (°C)), and the surface radiation temperature pattern (CP_sr) were explored. We executed two-way rmANOVA and Spearman's rho correlation analyses. Across all IRT parameters, mean CT_sr showed the highest association to cardiopulmonary parameters (E.g., oxygen consumption: r_s = -0.612 (running); -0.663 (cycling); p < .001). A global significant difference of CT_sr was identified between all relevant exercise test increments for both exercise-types (p < .001; η²_p = .842) and between both exercise-types (p = .045; η²_p = .205). Differences in CT_sr between running and cycling significantly appeared after a 3-min recovery period, whereas lactate, heart rate, and oxygen consumption were not different. High correlations between the CT_sr values extracted manually and the CT_sr values processed automatically by a deep neural network were identified. The applied objective time series analysis enables crucial insights into intra- and interindividual differences between both tests. CT_sr variations indicate different physiological demands between incremental running and cycling exercise testing. Further studies applying automatic ROI analyses are needed to enable the extensive analysis of inter- and intraindividual factors influencing the CT_sr variation during exercise to allow determine the criterion and predictive validity of IRT parameters in exercise physiology.

Keywords: Convolutional neural networks; Endurance exercise; Exercise radiomics; Infrared thermography; Thermal imaging.