Abstract:

This paper is devoted to a possible approach to aberrations correction in ultrasound images in case of heterogeneous media. Numerical modeling is performed for a direct problem - obtaining synthetic numerical ultrasound images using known geometry and rheology of the region, as well as transducer parameters. The numerical images reproduce distortions and artifacts that are typical for a medium containing areas with significantly different sound speeds. Convolutional neural networks are used to locate the interface of acoustically contrasting media. The located interfaces between areas with significantly different sound speeds are used to improve the quality of the image.
The results of the present work demonstrate that it is possible to improve the quality of images using reasonably fast algorithms based solely on information from the ultrasound sensor. The discussion section of the paper mentions the problems that should be addressed to allow future hardware implementation of the proposed approach.