Thermogram-based estimation of foot arterial blood flow using neural networks

The altered blood flow in the foot is an important indicator of early diabetic foot complications. However, it is challenging to measure the blood flow at the whole foot scale. This study presents an approach for estimating the foot arterial blood flow using the temperature distribution and an artificial neural network. To quantify the relationship between the blood flow and the temperature distribution, a bioheat transfer model of a voxel-meshed foot tissue with discrete blood vessels is established based on the computed tomography (CT) sequential images and the anatomical information of the vascular structure. In our model, the heat transfer from blood vessels and tissue and the inter-domain heat exchange between them are considered thoroughly, and the computed temperatures are consistent with the experimental results. Analytical data are then used to train a neural network to determine the foot arterial blood flow. The trained network is able to estimate the objective blood flow for various degrees of stenosis in multiple blood vessels with an accuracy rate of more than 90%. Compared with the Pennes bioheat transfer equation, this model fully describes intra- and inter-domain heat transfer in blood vessels and tissue, closely approximating physiological conditions. By introducing a vascular component to an inverse model, the blood flow itself, rather than blood perfusion, can be estimated, directly informing vascular health.