Non-Invasive Deep Temperature Measurement Based on the Long Short Term Memory for Hyperthermia Therapy

Abstract

In this study, we developed the model predicted the deep temperatures from the surface temperature information in order to realize non-invasive measurement for the hyperthermia therapy. The deep temperatures were predicted based on the surface temperature, surface temperature change, initial surface temperature, and lapsed time by using deep learning method based on long short term memory. The model was learned by using temperature characteristics measured by biological phantoms composed by agar. Errors of the model’s prediction accuracies for the phantoms were around 0.45 degree at the largest point. We measured the temperature characteristics of the pork-based phantom as a material similar to human tissue and used the model to make predictions. Errors of the prediction accuracies for the phantom were around 5.0 degree at the largest point. In this study, we used two type heat sources. The model does not enough learn temperature characteristics for each heat source. We confirmed that the system was able to achieve a prediction accuracy of less than 0.3 degree for data using a heat pack as a heat source