Prediction of laser-induced thermal damage with artificial neural networks

Abstract

Laser-induced thermotherapy (LITT) has been widely studied since it is a minimally invasive technique for focal destruction of liver tumors without side effects. However, there are still some concerns about the treatment and monitoring thermal effects during operation. Although real-time imaging modalities are available, like magnetic resonance imaging, they are not cost-effective and not applicable to all conditions. This paper presents artificial neural network (ANN) modeling of laser-induced thermal damages on ex vivo liver tissue. In this work three laser sources, i.e. a diode pumped laser with a wavelength of 980 nm, a 1070 nm yttrium lithium fluoride fiber laser, and a 1940 nm thulium fiber laser, were used in order to thermally damage tissues by applying the laser until coagulation observed. The diameter and depth of coagulation were empirically measured and used to train the ANN model by finding the correlation between laser parameters (application time, power, penetration depth, wavelength, and spot size) and thermal damages. Thermal damage can be determined by observing coagulation diameter and coagulation depth. The prediction ability and accuracy of the trained ANN model were tested by comparing the actual and simulated results. Our result showed that the ANN successfully predicts LITT damage in terms of coagulation diameter and coagulation depth, with a very high accuracy. The trained ANN model was compared with two mathematical models. In terms of performance, the ANN is a very useful and practical tool for determining LITT damage.