Efficient B-Mode Ultrasound Image Reconstruction From Sub-Sampled RF Data Using Deep Learning.

Area of Science:

• Medical Imaging
• Ultrasound Technology
• Artificial Intelligence in Medicine

Background:

• High-quality ultrasound imaging requires numerous radio-frequency (RF) measurements.
• Receiver (Rx) or transmit (Xmit) event sub-sampling in ultrasound systems leads to RF data limitations.
• Standard beamformers produce blurry images with artifacts due to RF sub-sampling, hindering diagnostic use.

Purpose of the Study:

• To develop a novel deep learning approach for reconstructing high-quality ultrasound images from limited RF measurements.
• To address the limitations of existing compressed sensing methods in ultrasound image reconstruction.
• To reduce data rates in ultrasound systems without sacrificing image quality.

Main Methods:

• A novel deep learning approach was proposed to interpolate missing RF data.
• The method utilizes redundancy in the Rx-Xmit plane for data interpolation.
• The approach was tested using sub-sampled RF data from a multi-line acquisition B-mode system.

Main Results:

• The proposed deep learning method effectively interpolates missing RF data.
• Experimental results confirmed the method's ability to reduce data rates.
• Image quality was maintained despite the reduction in RF measurements.

Conclusions:

• The novel deep learning approach successfully reconstructs high-quality ultrasound images from sub-sampled RF data.
• This method offers an effective solution for reducing data rates in ultrasound imaging systems.
• The technique shows promise for improving diagnostic capabilities in portable, 3-D, and ultra-fast ultrasound applications.