Improved DTI-Based Skeletal Muscle Architecture Estimation via Diffusion Weighted Image Denoising and First Eigenvector Map Smoothing.

Skeletal muscle architecture-the internal arrangement of a muscle's fibers with respect to its line of action-is a major determinant of muscle function. In diffusion tensor imaging (DTI) tractography, the first eigenvector of the diffusion tensor is integrated to create "fiber-tracts" that represent muscle architecture at the spatial scale of several fascicles. However, noise contamination of the images causes erroneous estimates of the first eigenvector that propagate during tractography, causing inaccurate architecture estimates at typical signal-to-noise ratios (SNR's). Although image denoising is commonly used, its effects have not been evaluated in skeletal muscle against known ground truth. Moreover, alternative strategies (such as smoothing of the muscle's first eigenvector field) remain unexplored. Therefore, simulated diffusion tensor images of a model muscle were used to quantify the effect of anisotropic image smoothing, threshold principal component analysis-based image denoising, and first eigenvector field smoothing on the accuracy of DTI-based muscle architecture estimates at different SNR levels. The denoising methods were then implemented in a human dataset. In the simulated dataset, anisotropic image smoothing and first eigenvector field smoothing reduced the deviation of the first eigenvector obtained from the noise-contaminated images from those obtained from the noise-free images and improved the accuracy of the fascicle curvature estimates. In the human dataset, both smoothing methods decreased the fiber-tract curvature estimates compared to the values in the raw images. Both anisotropic image smoothing and smoothing of the first eigenvector field improve the accuracy of DTI tractography-based muscle architecture estimates.