Reconstruction for Diverging-Wave Imaging Using Deep Convolutional Neural Networks.

Area of Science:

• Medical Imaging
• Ultrasound Technology
• Artificial Intelligence in Medicine

Background:

• Diverging wave (DW) ultrasound imaging offers high temporal resolution for cardiovascular applications.
• Limited DW transmissions result in lower image quality compared to focused ultrasound.
• Conventional reconstruction methods are computationally intensive and time-consuming.

Purpose of the Study:

• To develop a high-quality reconstruction method for DW ultrasound images using a reduced number of transmissions.
• To address the limitations of conventional reconstruction techniques in DW imaging.
• To improve image quality and reduce computational load in cardiovascular ultrasound.

Main Methods:

• Proposed a convolutional neural network (CNN) architecture named Inception for DW Network (IDNet).
• Utilized inception modules with multiscale convolution kernels to capture diverse image features.
• Trained the network using in vitro and in vivo samples to map low-quality to high-quality images.

Main Results:

• IDNet achieved high-quality image reconstruction using only 3 DW transmissions.
• Image quality was comparable to conventional compounding of 31 DWs.
• Outperformed standard compounding and conventional CNN methods in contrast ratio, contrast-to-noise ratio, and lateral resolution.

Conclusions:

• IDNet significantly enhances DW ultrasound image quality with minimal data.
• The proposed CNN architecture offers improved efficiency in terms of complexity and inference time.
• IDNet represents a promising advancement for real-time cardiovascular imaging.