Fast Strain Estimation and Frame Selection in Ultrasound Elastography Using Machine Learning.

Area of Science:

• Medical Imaging
• Biophysics
• Signal Processing

Background:

• Ultrasound elastography estimates tissue mechanical properties by analyzing deformation from ultrasound radio frequency (RF) signals.
• Time delay estimation (TDE) is crucial for calculating tissue displacement but faces challenges with computational complexity and selecting appropriate RF frames.
• Inadequate RF frame pairs can lead to unreliable strain estimation due to decorrelation or insufficient deformation.

Purpose of the Study:

• To develop a computationally efficient method for estimating tissue displacement in quasi-static ultrasound elastography.
• To introduce a novel approach for assessing the suitability of ultrasound RF frame pairs for accurate strain estimation.
• To improve the speed and reliability of ultrasound elastography for mechanical property assessment.

Main Methods:

• Principal Component Analysis (PCA) was used to learn 12 displacement modes from a large database of displacement fields.
• Dynamic Programming (DP) was employed for an initial, sparse displacement estimate, which was then decomposed into the learned PCA modes.
• The Global Ultrasound Elastography (GLUE) method was used to refine the displacement estimate, creating the PCA-GLUE method.
• A Multilayer Perceptron (MLP) classifier was trained using the PCA-GLUE weight vector to predict RF frame pair suitability for strain estimation.

Main Results:

• The PCA-GLUE method achieved over 10x speed improvement compared to DP for initial displacement map calculation, yielding identical results.
• The MLP classifier demonstrated high accuracy in determining RF frame pair suitability, achieving an F1-measure over 92% with a testing time of 1.5 ms.
• Validation was successfully performed using simulated, phantom, and in vivo ultrasound data.

Conclusions:

• PCA-GLUE significantly accelerates the initial displacement estimation in ultrasound elastography by reducing computational complexity.
• The developed MLP classifier effectively identifies suitable RF frame pairs, enhancing the reliability of strain estimation.
• This combined approach offers a faster and more robust solution for quantitative ultrasound elastography applications.