RECONSTRUCTION AND SEGMENTATION OF PARALLEL MR DATA USING IMAGE DOMAIN DEEP-SLR.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neuroscience

Background:

• Parallel MRI (PMRI) data acquisition often involves undersampling, leading to artifacts.
• Accurate reconstruction and segmentation are crucial for PMRI brain data analysis.
• Existing methods for uncalibrated PMRI recovery have limitations.

Purpose of the Study:

• To develop a novel framework for joint reconstruction and segmentation of undersampled PMRI brain data.
• To introduce an image domain deep network for calibrationless recovery of PMRI data.
• To minimize segmentation errors caused by undersampling artifacts and improve reconstruction quality.

Main Methods:

• Developed an image domain deep learning network for calibrationless recovery of undersampled PMRI data.
• Generalized local low-rank based approaches for uncalibrated PMRI recovery.
• Combined the reconstruction scheme with a segmentation network, trained end-to-end.

Main Results:

• The proposed deep learning approach demonstrated improved performance in both reconstruction and segmentation.
• End-to-end training minimized segmentation errors resulting from undersampling artifacts.
• Reconstructed images exhibited reduced blurring and sharper edges compared to independently trained networks.

Conclusions:

• The novel framework offers a robust solution for joint reconstruction and segmentation of PMRI brain data.
• The image domain deep network approach shows promise for enhanced PMRI data analysis.
• This method improves image quality and reduces segmentation errors, advancing PMRI applications.