Synthetic Elastography Using B-Mode Ultrasound Through a Deep Fully Convolutional Neural Network.

Area of Science:

• Biomedical Imaging
• Medical Ultrasound
• Artificial Intelligence in Medicine

Background:

• Shear-wave elastography (SWE) is a key technique for measuring tissue elasticity.
• Estimating tissue elasticity is crucial for diagnosing various medical conditions.
• Current SWE requires specialized equipment and operator expertise.

Purpose of the Study:

• To develop a deep learning method for generating synthetic SWE (sSWE) images from standard B-mode ultrasound.
• To assess the accuracy of sSWE in estimating tissue elasticity.
• To explore the potential of sSWE for broader clinical applications.

Main Methods:

• A deep learning model was trained using paired B-mode and SWE images from 50 prostate cancer patients.
• Synthetic SWE images were generated from B-mode images.
• The accuracy of sSWE was evaluated by comparing it to original SWE images using mean absolute error.
• t-distributed stochastic neighbor embedding was used for visualizing feature-level data.

Main Results:

• sSWE images were generated with a pixel-wise mean absolute error of 4.5 ± 0.96 kPa compared to original SWE.
• The method demonstrated good overlap between data from different ultrasound scanners.
• sSWE was qualitatively assessed on B-mode images from a scanner lacking SWE functionality.
• The approach was also tested for elasticity imaging in the thyroid.

Conclusions:

• Deep learning enables the generation of synthetic SWE images from conventional B-mode ultrasound.
• sSWE shows promise as a tool for estimating tissue elasticity, potentially overcoming limitations of conventional SWE.
• Further research is needed to standardize the technique and validate its clinical utility across different organs.