Intelligent non-invasive modeling of ultrasound-induced temperature in tissue phantoms

Raising temperature of human cells (hyperthermia) is an ancient tool for tumor masses reduction and extinction, actually even before the existence of a molecular understanding of cancer cells. Hyperthermia is being increasingly used for patients' rehabilitation and oncological diseases' treatment but still constitutes a major driver for researching more efficient and reliable therapeutic usage aiming at outstanding patients wellbeing and socio-economic benefits. Efficient hyperthermia practice demands knowledge about the exact amount of heating required at a particular tissue location, as well as information concerning the spatial heating distribution. Both of these processes require accurate characterization. Until now, ultrasound heating treatments are being monitored by magnetic resonance imaging (MRI), recognized as being capable of achieving a 0.5 °C/cm3 temperature resolution [1], thereby imposing a gold standard in this field. However, one can notice that MRI-based techniques, besides the inconvenient instrumental cost, obliges the presence of a team of expert clinicians and limits the hyperthermia ultrasound treatment area due to the space restrictions of an MRI examination procedure. This article introduces a novel non-invasive modelling approach of ultrasound-induced temperature propagation in tissues, to be used as a cost effective alternative to MRI monitoring of ultrasound therapeutic techniques, achieving a maximum temperature resolution of 0.26 °C/cm3, clearly inferior to the MRI gold standard resolution of 0.5 °C/cm3. In order to derive the model, and avoiding painful invasive in-vivo sampling, a phantom was employed, whose composition respects the human tissues' reaction to ultrasound beams. In contrast with previous works of the authors, in the present paper we study the possibility of using b-spline neural networks (BSNN) as reliable noninvasive estimator of temperature propagation in phantoms [2,3]. The proposed methodology achieves better results than previous approaches, does not require the use of an Imaging Ultrasound transducer and, as the proposed models are piecewise polynomial models, they can be easily inverted and used in closed-loop control of therapeutic ultrasound instruments.