Improved multi-echo gradient echo myelin water fraction mapping using complex-valued neural network analysis.

Area of Science:

• Neuroimaging
• Biomedical Engineering
• Artificial Intelligence

Background:

• Myelin water fraction (MWF) is a key MRI biomarker for white matter integrity.
• Existing artificial neural network (ANN) methods for MWF estimation can be biased by white matter fiber orientation.
• Accurate MWF quantification is crucial for diagnosing and monitoring neurological disorders.

Purpose of the Study:

• To develop an advanced workflow for improved quantitative myelin water fraction (MWF) estimation.
• To address and correct biases in MWF quantification caused by white matter fiber orientation.
• To enhance the reliability and confidence of MWF mapping in neuroimaging.

Main Methods:

• Utilized a complex-valued neural network (CVNN) with complex-valued operations to account for fiber orientation effects.
• Developed a signal model incorporating T1 values for training data generation to compensate for scan parameter variations.
• Implemented a voxel-spread function for spatial B0 artifact correction and dropout-based variational inference for uncertainty estimation.

Main Results:

• The proposed CVNN method demonstrated improved MWF estimation quality compared to previous ANN methods.
• The approach effectively corrected for biases related to fiber orientation and spatial artifacts.
• Uncertainty estimates provided a confidence level for the resulting MWF values, distinct from fitting error.

Conclusions:

• An improved method for myelin water fraction (MWF) mapping has been developed using complex-valued neural network analysis.
• The advanced workflow enhances the accuracy and reliability of MWF quantification in neuroimaging.
• This technique offers a more robust approach to assessing white matter integrity.