Cardiac MR image reconstruction using cascaded hybrid dual domain deep learning framework.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Biomedical Engineering

Background:

• Cardiac MRI image reconstruction from under-sampled data is challenging due to motion artifacts.
• Compressed Sensing (CS) and Parallel Imaging (pMRI) offer acceleration but have limitations in high-resolution imaging.
• Deep learning (DL) methods show promise but single-domain approaches don't fully leverage image and k-space correlations.

Purpose of the Study:

• To develop and evaluate a dual-domain deep learning approach for enhanced cardiac MRI reconstruction.
• To address limitations of conventional and single-domain DL methods in handling under-sampled data.
• To improve image quality, reduce artifacts, and increase reconstruction accuracy.

Main Methods:

• A hybrid dual-domain deep learning model integrating image and k-space domains was developed.
• The model incorporates multi-coil data consistency (MCDC) layers.
• Reconstruction was performed on 1-D Variable Density (VD) random under-sampled cardiac MRI data.

Main Results:

• The proposed dual-domain DL method demonstrated superior performance over conventional DL and CS techniques.
• Quantitative metrics showed higher Structural Similarity Index (SSIM) and Peak Signal-to-Noise Ratio (PSNR).
• Lower Root Mean Square Error (RMSE) indicated improved accuracy and reduced artifacts.

Conclusions:

• The hybrid dual-domain deep learning approach with MCDC layers effectively reconstructs diagnostic-quality cardiac MR images.
• This method offers enhanced robustness and accuracy for accelerated cardiac MRI.
• The findings suggest a promising direction for advanced medical image reconstruction.