Rapid machine-learning enabled design and control of precise next-generation cryogenic surgery in dermatology

Abstract

In the field of dermatology, the use of cryogenic processes, such as cryoablation, cryotherapy, etc., have grown dramatically over the last decade. This usually entails using a cryoprobe to freeze and destroy unwanted tissue, such as cancer cells. The focus of this work is to develop a digital-twin (a digital replica) of the performance of a cryogenic probe, which can be used to pre-plan and optimize surgical procedures, in order to maximize successful outcomes. Specifically, we model the optimal cryoprobe-induced cooling protocol needed to eliminate cells/tissue in specific regions, while minimizing damage to nearby tissue. The modeling approach is to develop mathematical surface point-source heat extraction kernels and then to create optimal surface patterns that the cryoprobe induces, by arranging the point-sources accordingly. Spatial and temporal control of the heat extraction is modeled. The entire subdermal thermal field is then constructed by superposing the solutions, enabling precise cryogenic treatment. Finally, a Machine Learning Algorithm (MLA) is then applied to optimize the set of parameters to deliver a precise response, making it an ideal real-time surgical tool.