Machine Learning for H-FIRE Protocols: Tuning Parameters for High-Frequency Irreversible Electroporation by Machine Learning

Abstract

Microwave (MW) technologies are increasingly being employed in biomedicine and health care. Applications such as MW ablation (MWA), MW sensors for the realtime monitoring of physiological parameters, and lab-on-a-chip devices are just some of the examples in a wide variety of possibilities. Recently, high-frequency irreversible electroporation (H-FIRE) has been attracting more attention due to its important applications in oncology. It uses pulsed electrical fields (PEFs) to induce a controlled but irreversible process of permeabilization of cell membranes, thus triggering a substantial alteration in the physiological equilibria of cells, ultimately leading to their death. In this article we proposed an ML-based approach that exploits an ANN to estimate ablation occurrence, ablation area, and E-field lethal threshold (i.e., the three parameters highly impacting each H-FIRE therapeutic application). The ANN was modeled and trained by examining a large set of scientific publications describing H-FIRE experiments and their outcomes. Starting from these data, the ANN was capable of providing a satisfactory estimation accuracy on the aforementioned parameters. The achieved results, although very promising, could be further improved by widening the initial data set (i.e., by gathering details about a larger number of H-FIRE experiments) as well as by evaluating more complex ANN models. The presented tool has powerful potential to achieve fast, personalized protocols in H-FIRE therapy, streamlining the long and complex treatment planning process based on standard electromagnetic and multiphysics simulation methods for a new era of H-FIRE applications.