Ultrafast diffusion tensor imaging based on deep learning and multi-slice information sharing.

Area of Science:

• Medical Imaging
• Neuroimaging
• Biophysics

Background:

• Diffusion Tensor Imaging (DTI) quantifies tissue microstructure non-invasively.
• Traditional DTI requires numerous diffusion directions, leading to long scan times and motion sensitivity.
• Reducing scan time without compromising DTI reconstruction quality is crucial for clinical applications.

Purpose of the Study:

• To develop a fast DTI acquisition scheme using fewer diffusion directions.
• To implement a deep learning-based reconstruction method for high-quality DTI.
• To enable rapid DTI acquisition for wider clinical adoption.

Main Methods:

• A novel DTI scan scheme utilizing three diffusion directions per slice with a specific switching pattern.
• A deep learning reconstruction method employing multi-slice information sharing and T1-weighted images.
• A U-Net-based network with two encoders for efficient diffusion data utilization and direct nonlinear mapping to diffusion tensors.

Main Results:

• High-quality mean diffusivity, fractional anisotropy, and directionally encoded colormaps were achieved with only three diffusion directions per slice.
• The method demonstrated robust performance on both Human Connectome Project and clinical patient data.
• DTI-derived maps were successfully reconstructed in under one minute of scan time.

Conclusions:

• The proposed fast DTI method significantly reduces scan time while preserving reconstruction quality.
• This advancement facilitates the broader clinical application of DTI for microstructure assessment.
• Rapid DTI acquisition holds promise for improved diagnostic capabilities and patient throughput.