Mitigating transmit-B₁ artifacts by predicting parallel transmission images with deep learning: A feasibility study...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neuroscience

Background:

• B1+ artifacts are a common issue in MRI, particularly at high field strengths like 7T.
• Parallel transmission (pTx) is often used to mitigate these artifacts but requires specialized hardware.
• Single-channel transmission (sTx) is more widely accessible but prone to B1+ related signal dropout.

Purpose of the Study:

• To develop a novel deep learning (DL) approach for B1+ artifact mitigation.
• To predict parallel transmission (pTx) images from single-channel transmission (sTx) images, bypassing the need for pTx hardware.
• To enhance image quality and downstream analysis in diffusion MRI.

Main Methods:

• A deep encoder-decoder convolutional neural network was designed and trained to map sTx to pTx images.
• The model was trained and validated on 7T Human Connectome Project (HCP)-style diffusion MRI data from healthy subjects.
• Hyperparameter tuning and generalization performance were assessed using nested and regular cross-validation, respectively.

Main Results:

• The DL method successfully restored signal dropout in sTx images, improving key quality metrics (NRMSE, PSNR, SSIM).
• Image quality improvements translated to enhanced diffusion tensor imaging analysis, including more accurate fractional anisotropy and mean diffusivity estimations.
• The model demonstrated effective generalization to unseen subjects, recovering signal dropout and reducing noise in fractional anisotropy maps.

Conclusions:

• The proposed DL method offers a viable solution for B1+ artifact reduction in sTx MRI.
• This approach can improve image quality and analytical performance, especially when pTx is not accessible.
• The technique shows promise for routine clinical application in healthy subjects.