Accelerating CEST MRI With Deep Learning-Based Frequency Selection and Parameter Estimation.

Area of Science:

• Biomedical Imaging
• Machine Learning in Medical Imaging
• Molecular Imaging

Background:

• Chemical Exchange Saturation Transfer (CEST) MRI is sensitive for metabolite detection but limited by long scan times.
• Prolonged acquisition in CEST MRI is due to the extensive frequency offsets required for parameter estimation.
• Reducing frequency offsets is key to accelerating CEST MRI acquisition.

Purpose of the Study:

• To develop and validate a deep learning framework for accelerating CEST MRI.
• To integrate frequency selection and parameter estimation to reduce scan time.
• To assess the performance of the proposed method compared to existing techniques.

Main Methods:

• A novel deep learning framework utilizing channel pruning via batch normalization for informative frequency selection.
• Simultaneous training of the network for accurate parametric map prediction (APT, NOE, MT).
• Reconstruction using MR Multitasking with a low-rank tensor model for k-space under-sampling.

Main Results:

• The framework identified 13 informative frequency offsets from 53, significantly reducing acquisition time.
• Deep learning-based parametric maps were comparable in quality to those from all offsets.
• Achieved a whole-brain CEST MRI scan time reduction from 5:30 min to under 1:30 min.
• Outperformed previous Fisher information-based selection methods.

Conclusions:

• The proposed deep learning framework effectively accelerates CEST MRI by intelligent frequency selection.
• This method significantly reduces scan time without compromising diagnostic image quality.
• The approach shows promise for efficient and practical clinical implementation of CEST MRI.