PROCESSAMENTO DE SINAIS ULTRASÓNICOS DE HIPERTERMIA E CONTROLO DE SISTEMA TERAPÊUTICO

Funder

Organizational Unit

Publications

Soft-computing techniques applied to artificial tissue temperature estimation

Publication . Teixeira, C. A.; Ruano, M. Graca; Ruano, A. E.

Safety and efficiency of thermal therapies strongly rely on the ability to quantify temperature evolution in the treatment region. Research has been developed in this field, and both invasive and non-invasive technologies have been reported. Till now, only the magnetic resonance imaging (MRI) achieved the hyperthermia/diathermia gold standard value of temperature resolution of 0.5oC in 1cm3, in an in-vivo scenario. However, besides the cost of MRI technology, it does not enable a broad-range therapy application due to its complex environment. Alternatively, backscattered ultrasound (BSU) seems a promising tool for thermal therapy, but till now its performance was only quantitatively tested on homogeneous media and on single-intensity and three-point assessment have been reported. This thesis reports the research performed on the evaluation of time-spatialtemperature evolution based mainly on BSU signals within artificial tissues. Extensive operating conditions were tested on several experimental setups based on dedicated phantoms. Four and eight clinical ultrasound intensities, up to five spatial points, homogeneous and heterogeneous multi-layered phantoms were considered. Spectral and temporal temperature-dependent BSU features were extracted, and applied as invasive and non-invasive methodologies input information. Softcomputing methodologies have been used for temperature estimation. From linear iterative model structure models, to multi-objective genetic algorithms (MOGA) model structure optimisation for linear models, radial basis functions neural netxi xii works (RBFNNs), RBFNNs with linear inputs (RBFLICs), and for the adaptivenetwork- based fuzzy inference system (ANFIS) have been used to estimate the temperature induced on the phantoms. The MOGA+RBFNN methodology, fed with completely data-driven information, estimated temperature with maximum absolute errors less than 0.5oC within two spatial axes. The proposed MOGA+RBFNN methodology applied to non-invasive estimation on multi-layered media, is a innovative approach, as far as known, and enabled a step forward on the therapeutic temperature characterisation, motivating future instrumentation temperature control.

2008Doctoral thesis

Open Access