Automatic Frame Selection using CNN in Ultrasound Elastography.

Area of Science:

• Medical imaging
• Biomedical engineering
• Artificial intelligence in medicine

Background:

• Ultrasound elastography estimates tissue mechanical properties by analyzing deformation response to applied forces.
• Tissue stiffness influences deformation; stiffer tissues exhibit less deformation.
• Strain image quality in elastography is critically dependent on motion type, with in-plane axial motion yielding superior results compared to out-of-plane motion.

Purpose of the Study:

• To introduce a novel convolutional neural network (CNN) for rapid assessment of radio frequency (RF) frame pair suitability for ultrasound elastography.
• To enable automatic selection of optimal RF frame pairs for generating high-quality strain images.
• To significantly reduce the processing time for elastography data analysis.

Main Methods:

• Development and implementation of a CNN model to evaluate RF frame pairs for elastography.
• Training the CNN on a dataset of 3,818 RF frame pairs from both phantom and in vivo data.
• Testing the CNN on 986 unseen RF frame pairs to validate its performance.

Main Results:

• The CNN achieved an accuracy exceeding 91% in determining RF frame pair suitability for elastography.
• The method demonstrated a processing time of only 5.4 milliseconds per frame pair.
• Successful application on both phantom and in vivo ultrasound data.

Conclusions:

• The proposed CNN method offers a fast and accurate solution for selecting suitable RF frame pairs in ultrasound elastography.
• This approach has the potential to automate the process of identifying optimal data for high-quality strain imaging.
• The findings suggest a significant advancement in improving the reliability and efficiency of ultrasound elastography analysis.