Transformer Enabled Half Z‑spectrum Sampling B₀ Inhomogeneity Correction for GluCEST and NOE MRI.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Biophysics

Background:

• Chemical Exchange Saturation Transfer (CEST) MRI is crucial for quantifying metabolites like glutamate.
• B0 inhomogeneity causes significant quantification errors in CEST MRI, necessitating correction.
• Existing deep learning methods reduce scan time but can be further optimized.

Purpose of the Study:

• To develop a Transformer-based model for efficient B0 inhomogeneity correction in Glutamate-weighted CEST (GluCEST) and NOE MRI.
• To reduce scan time by approximately 50% compared to previous deep learning approaches.
• To achieve superior B0 correction accuracy using a reduced set of Z-spectrum offset images.

Main Methods:

• Developed distinct Swin Transformer networks for positive and negative Z-spectrum sides.
• Trained networks to map limited GluCEST images to specific 3 ppm peaks.
• Applied a similar methodology to NOE CEST imaging, optimizing for spectral characteristics.

Main Results:

• Transformer models significantly outperformed previous deep learning methods visually and quantitatively.
• Achieved a 50% reduction in data acquisition time by using only positive Z-spectrum offsets.
• Maintained high B0 inhomogeneity correction accuracy with the reduced dataset.

Conclusions:

• Efficient B0 inhomogeneity correction in GluCEST and NOE MRI is feasible using select downfield Z-spectrum offset images.
• Acquisition time can be reduced by over 80% with this approach.
• The Transformer-based model offers a robust, efficient, and superior alternative to CNNs for clinical CEST MRI.