Ultrasound tomography enhancement by signal feature extraction with modular machine learning method

Area of Science:

• Engineering
• Computer Science
• Materials Science

Background:

• Industrial and general ultrasound tomography requires robust diagnostic methods.
• Current object detection techniques may lack reliability or incur high costs.

Purpose of the Study:

• To present a novel deep neural network approach for object detection in ultrasound tomography.
• To analyze time-dependent ultrasonic signals for common signal characteristics.

Main Methods:

• Utilized custom-designed deep neural network models with convolution and dense layers.
• Analyzed time-dependent ultrasonic signals from three independent transducers.
• Focused on extracting common signal features for object property prediction.

Main Results:

• Achieved high accuracy in predicting the number of objects (R2 = 99.8%).
• Demonstrated high accuracy in predicting object locations and sizes (R2 = 98.4%).
• The deep neural network effectively identified common signal characteristics.

Conclusions:

• The proposed deep neural network offers a reliable and low-cost method for object detection in ultrasound tomography.
• This approach has significant potential for application across various industry sectors.
• The method effectively leverages signal analysis for accurate object characterization.