Artificial neural network for myelin water imaging

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Medical Image Analysis

Background:

• Myelin water imaging (MWI) is crucial for assessing white matter integrity in neurological diseases.
• Current MWI processing is computationally intensive, limiting real-time applications.
• Artificial neural networks (ANNs) offer potential for accelerating complex image analysis tasks.

Purpose of the Study:

• To develop and validate ANNs for rapid, accurate MWI data processing.
• To demonstrate the feasibility of real-time MWI analysis using deep learning.
• To compare ANN-derived MWI parameters with conventional methods.

Main Methods:

• Three ANNs (ANN-I_MWF, ANN-I_GMT2, ANN-II) were trained and tested using gradient and spin echo data from healthy controls and multiple sclerosis patients.
• Networks were designed to output myelin water fraction (MWF) and geometric mean T₂ (GMT_2,IEW), or T₂ distribution.
• ANN performance was evaluated against conventional MWI using normalized root-mean-squared error and statistical comparisons, including analysis of scan parameter effects.

Main Results:

• ANNs achieved high accuracy, with averaged normalized root-mean-squared errors below 3% for MWF and 0.4% for GMT_2,IEW on the test set.
• Differences between ANN and conventional MWI were minimal (<0.1% for MWF, <0.1 ms for GMT_2,IEW) with high correlation (R² > 0.97).
• ANN processing achieved a 11,702-fold acceleration (0.68 s vs. 7,958 s) compared to conventional MWI.

Conclusions:

• The developed ANNs demonstrate the feasibility of real-time MWI processing.
• ANNs provide a highly accurate and significantly accelerated alternative for MWI analysis.
• This approach holds promise for improving the clinical utility of MWI in diagnosing and monitoring neurological disorders.