SMC-DTI: Simultaneous Motion Correction and DTI parameter estimation Via unsupervised deep learning framework

Motivation: Diffusion Tensor Imaging (DTI) provides insights into brain microstructure and connectivity, but acquiring sufficient diffusion-weighted images (DWI) for accurate estimation requires lengthy scans, increasing the risk of misalignment from patient motion between acquisitions. Goal(s): This work aims to streamline the DTI analysis process by integrating registration and parameter estimation into a unified framework, where both tasks are co-dependent, eliminating the need for separate, time-consuming steps. Approach: We present SMC-DTI, a deep-learning framework that jointly optimizes registration and DTI estimation by leveraging motion correction and DTI fitting interdependence. Results: Through comparisons with traditional and deep learning-based DTI analysis methods, SMC-DTI demonstrated superior performance. Impact: This study enables accurate brain microstructure estimation from DTI data with limited gradient directions affected by motion between acquisitions, reducing patient discomfort, improving subject experience, and potentially increasing imaging center throughput by shortening scan times.