Subcutaneous temperature prediction during cryogenic jet cooling by surface temperature measurements and RNNs

Abstract

Accurate information of subcutaneous temperature is crucial to the effectiveness and safety of a variety of medical procedures by energy-based devices, such as cooling treatments. Numerous strategies for invasive subcutaneous temperature monitoring have been investigated; however, they are impractical mainly due to their requirements for precise heat transfer modeling. In the case of cooling treatment by cryogenic substances, heat transfer rate largely varies with their phase change and procedure condition, which prevents accurate prediction of subcutaneous temperature. In this work, we propose a non-invasive, accurate, and practical prediction of subcutaneous temperature during CO2 spray jet cooling without numerical modeling by utilizing recurrent neural networks (RNNs) with surface temperature measurements for the first time. A patchable temperature sensor array on a flexible PCB is used to measure surface temperature. With this measured surface temperature, RNNs are developed with two functionalities: extraction of thermophysical properties (thermal conductivity and thermal diffusivity) of the target substrate and prediction of temperature response at 1-mm depth from the surface, which is the typical location of pain points. Five polymer thin films are used with the reference thermocouple embedded inside to train RNNs. A porcine skin is used to validate the RNNs, which show a prediction accuracy of 99.2%.