Magnetic Resonance Spectroscopy Spectral Registration Using Deep Learning.

Area of Science:

• Magnetic Resonance Spectroscopy (MRS)
• Deep Learning Applications
• Medical Imaging Analysis

Background:

• Deep learning excels in MRI image registration but lacks methods for MRS spectral registration (SR).
• Accurate spectral registration is crucial for quantitative MRS analysis.

Purpose of the Study:

• To develop and evaluate a convolutional neural network-based SR (CNN-SR) approach.
• To achieve simultaneous frequency-and-phase correction (FPC) for single-voxel MEGA-PRESS MRS data.

Main Methods:

• Utilized 40,000 simulated and 101 in vivo MEGA-PRESS datasets.
• Trained and validated a CNN-SR model on simulated data with varying offsets and SNRs.
• Assessed performance using mean absolute errors for simulated data and choline interval variance for in vivo data.

Main Results:

• CNN-SR accurately corrected frequency and phase offsets in simulated data across different SNRs.
• Achieved high performance in correcting offsets in in vivo MEGA-PRESS data.
• Demonstrated robust performance with small, medium, and large additional offsets.

Conclusions:

• The CNN-SR method provides an efficient and accurate solution for simultaneous FPC.
• This approach significantly improves spectral registration for MEGA-PRESS MRS data.
• CNN-SR holds promise for enhancing quantitative analysis in MRS studies.