The Effects of Deep Learning Imaging Denoising on Quantitative Diffusion Metrics of Lower Leg Muscles

Motivation: Deep learning (DL) denoising reconstruction reduces noise without added scan times. However, the effects of this denoising on quantitative diffusion tensor imaging (DTI) metrics have yet to be evaluated. Goal(s): Determine the effects of DL-denoising, across a range of signal averages (NEX), on quantitative DTI metrics in lower leg muscles. Approach: 6 Subjects obtained a series of DTI acquisitions. Quantitative biometrics of muscle compartments were compared with and without DL-denoising. Results: Biometric accuracy for each NEX was similar with and without DL-denoising. All coefficients of variation fell below 5%. Concordance correlation coefficients were above 0.87. Impact: The influence of DL-denoising on quantitative diffusion tensor imaging (DTI) metrics is unknown. We found that DL-denoising in lower leg muscles does not affect quantitative DTI measures but also does not improve accuracy at lower signal averages.