Spherical Harmonics-Based Deep Learning Achieves Generalized and Accurate Diffusion Tensor Imaging.

Area of Science:

• Medical Imaging
• Neuroscience
• Machine Learning

Background:

• Diffusion Tensor Imaging (DTI) is crucial for clinical and neuroscience research using MRI.
• Low signal-to-noise ratio in diffusion-weighted images compromises DTI reliability.
• Current deep learning (DL) methods for DTI lack generalization across different acquisition schemes.

Purpose of the Study:

• To develop a generalized, accurate, and efficient DL-based DTI method.
• To improve DTI quality and reliability in diverse clinical and research settings.
• To overcome the generalization limitations of existing DL approaches in DTI.

Main Methods:

• Representing diffusion MRI signals using spherical harmonics (SH) coefficients.
• Utilizing SH coefficient maps as input to a neural network for diffusion tensor (DT) field prediction.
• Validating the method on simulated and in-vivo datasets across various DTI scenarios.

Main Results:

• The proposed SH-DTI method demonstrated superior performance in quantitative and qualitative DTI analyses.
• Achieved advanced accuracy and efficiency in diffusion tensor field prediction.
• Showcased significant generalization capabilities across different acquisition schemes, centers, and scanners.

Conclusions:

• The novel SH-DTI method offers a generalized and robust solution for improving DTI quality.
• This approach enhances the broad applicability of DL in diverse neuroimaging settings.
• The method holds promise for more reliable clinical and neuroscience applications of DTI.