Applications of neural networks to ultrasound tomography

Ultrasound tomography holds promise in the area of medical diagnosis but is limited by the inadequacies of current reconstruction algorithms. An alternative method using neural networks is presented. The theory begins with X-ray tomography and is extended to ultrasound. After theoretical relevance is introduced, several experiments are discussed to illustrate the effectiveness of neural networks. The model was a circular cylinder with acoustic properties of tissue, insonated by a line source at 2 MHz. The transducers were arranged in a ring surrounding the cylinder, with one being the transmitter. The experiments involved varying the acoustic speed and the radius of the cylinder. In both cases, the neural network was able to generalize to parameters other than the ones used during the training.