Automatic Reconstruction Algorithms for In Vivo Skeletal Muscle 3D-SWEI Data

Rotational 3-D shear wave elasticity imaging (3D-SWEI) offers many opportunities for the development of biomechanical biomarkers of muscle health. However, methods for material parameter reconstruction to date have involved manual processing steps, which are impractical for clinical implementations. In this work, we describe and optimize algorithms for the automatic processing of 3D-SWEI skeletal muscle data and explore the most important factors affecting the success of automatic estimation of the shear wave speed (SWS) along and across the fibers. We compare the results of several automatic algorithms to the results obtained using manual processing across 130 in vivo 3D-SWEI datasets, evaluating the percent of valid acquisitions (non-gross-error), percent bias, and absolute percent error for each algorithm. We evaluate three major factors for algorithm optimization: wave detection method [comparing Radon sum, time to peak (TTP), and cross correlation (XCorr)], lateral range selection (comparing a single range for all acquisition angles, changing the range with fiber orientation, and dynamic range selection, in addition to analyzing the start and end positions of the lateral range), and methods to improve robustness (exploring detecting the wave through 3-D fits, and methods to isolate individual waves when multiple waves are detected). We conclude the optimal algorithm for passive muscle is a fixed lateral range (2-18 mm) Radon sum method with multiwave detection and SWS fitting to the ellipse predicted in transversely isotropic (TI) materials, and we provide results that can guide algorithm choice for other applications. In addition, we have made the implementation of all our algorithms publicly accessible at: DOI 10.5281/zenodo.18883176.