Real-time Control in Embedded Systems using Deep Neural Network-based Estimation Applied to Joule Heating


One method of cancer treatment is to thermally ablate (destroy) tumor masses using heat caused by electric current, or Joule heating. This particular modality is called radiofrequency ablation (RFA), due to the use of electric currents in the 100 kHz to 800 kHz frequency range. Computationally, RFA is simulated as a quasi-static Joule heating model within finite element analysis, composed of the computation of the solution to the Laplace partial differential equation for the electric problem and the bioheat equation for the thermal problem. This technique has yielded very accurate treatment planning and progress (measured by the depth or extent of tissue damage) estimation results, but the computation of finite element models is on the order of minutes and requires high-performance x64-based processors. Since thermal ablation is time-dependent, the more time spent on computation, the less time during the clinical procedure will be allocated to destroying potentially-cancerous tissue. Thus, we present a low-cost embedded system that allows for real-time control of radiofrequency ablation on ARMv7 microcontrollers. Deep neural networks are used to estimate the ablation depths in order to reach soft real-time computation. The primary contributions of this system are: (1) collection of sensor data for the RFA control process using only a single device within the target tissue, and (2) real-time transformation of the collected sensor data into data usable for RFA process control (ablation lesion depth).

Alternate Identifier
Date created
Resource type
Rights statement