Accelerating cardiac diffusion tensor imaging with deep learning-based tensor de-noising and breath hold reduction. A step towards improved efficiency and clinical feasibility.

BACKGROUND: Cardiac diffusion tensor imaging (cDTI) non-invasively provides unique insights into cardiac microstructure. Current protocols require multiple breath-hold repetitions to achieve an adequate signal-to-noise ratio, resulting in lengthy scan times. The aim of this study was to develop a cDTI de-noising method that would enable the reduction of repetitions while preserving image quality. METHODS: We present a novel de-noising framework for cDTI acceleration centered on three fundamental advances as follows: (1) a paradigm shift from image-based to tensor-space de-noising that better preserves structural information, (2) an ensemble of Vision Transformer-based models specifically optimized for tensor processing through adversarial training, and (3) a sophisticated data augmentation strategy that maximizes training data utilization through dynamic repetition selection. RESULTS: Our approach reduces scan times by a factor of up to 4 while achieving a 20% reduction in cDTI maps errors over existing de-noising methods (fractional anisotropy errors 0.09 vs 0.07) and preserving anatomical features such as infarct characterization and transmural cardiomyocyte orientation patterns. Crucially, our proposed method succeeds in clinical cases where other algorithms previously failed. CONCLUSION: This demonstrates substantial improvements in cDTI acquisition efficiency, achieving up to four-fold scan time reduction (3-5 breath-holds) while maintaining diagnostic accuracy across diverse cardiac pathologies.