Improving deep PROPELLER MRI via synthetic blade augmentation and enhanced generalization

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Magnetic Resonance Imaging

Background:

• PROPELLER MRI faces challenges in acquiring high-quality blade data, hindering deep learning reconstruction models.
• Standard data augmentation techniques for Cartesian MRI are not directly applicable to PROPELLER's unique blade geometry.

Purpose of the Study:

• To develop a novel method for generating synthetic PROPELLER blades for data augmentation.
• To improve the reconstruction of undersampled blades in PROPELLER MRI, enhancing image quality and computational efficiency.
• To increase the generalization capability and robustness of deep learning models for PROPELLER MRI.

Main Methods:

• Generation of synthetic PROPELLER blades.
• Application of synthetic blades for data augmentation in training deep learning reconstruction models.
• Evaluation using metrics such as Peak Signal-to-Noise Ratio (PSNR), Normalized Mean Square Error (NMSE), and Structural Similarity Index Measure (SSIM).

Main Results:

• Models trained with augmented data demonstrated superior performance compared to those trained without.
• Synthetic blade augmentation significantly improved the generalization capability of the deep learning model.
• The study confirmed the feasibility of training models exclusively on synthetic blades, reducing reliance on real data.

Conclusions:

• The novel synthetic blade generation and data augmentation technique enhances PROPELLER MRI reconstruction.
• This approach leads to improved image quality, computational efficiency, and model generalization.
• The findings suggest a reduced dependency on real PROPELLER MRI data for training deep learning models.