Neuro-genetic non-invasive temperature estimation: intensity and spatial prediction

Area of Science:

• Biomedical Engineering
• Computational Neuroscience

Background:

• Accurate non-invasive temperature estimation is crucial for effective thermal therapy.
• Existing methods require precise control of therapeutic instrumentation.

Purpose of the Study:

• To develop and validate radial basis functions artificial neural networks for predicting temperature evolution in ultrasound-insonated media.
• To assess the models' predictive capacity across various operational situations, including new spatial locations and intensity levels.

Main Methods:

• Radial basis functions neural networks with external dynamics were employed.
• A multi-objective genetic algorithm identified optimal model inputs and neuron count.
• Models were validated using both operational and unseen data, including new spatial and intensity parameters.

Main Results:

• The neuro-genetic models demonstrated good performance in predicting temperature propagation.
• A maximum absolute error of 0.5°C was achieved, meeting the gold standard for hyperthermia/diathermia.
• Low computational complexity models were developed.

Conclusions:

• The proposed neuro-genetic approach effectively predicts temperature changes based on intensity and space parameters.
• This method allows for the assessment of diverse operating conditions with adequate temperature resolution.
• The findings support enhanced control and precision in thermal therapy applications.