Seeking Common Ground While Reserving Differences: Multiple Anatomy Collaborative Framework for Undersampled MRI...

Area of Science:

• Medical Imaging
• Deep Learning
• Magnetic Resonance Imaging

Background:

• Current deep learning models for MRI reconstruction are anatomy-specific, leading to inefficiencies.
• These models overlook shared de-aliasing knowledge across different anatomies.
• Training a single network on all anatomies can degrade performance due to conflicting exclusive knowledge.

Purpose of the Study:

• To develop a novel deep MRI reconstruction framework that leverages both shared and anatomy-specific knowledge.
• To address the limitations of the one-anatomy-one-network approach in MRI reconstruction.
• To improve the efficiency and performance of undersampled MRI reconstruction.

Main Methods:

• Proposed a framework with anatomy-shared learners trained on diverse anatomies and anatomy-specific learners trained on target anatomies.
• Explored four different implementations of anatomy-specific learners within the framework.
• Applied the framework to two deep MRI reconstruction networks and evaluated on brain, knee, and cardiac MRI datasets.

Main Results:

• Demonstrated enhanced reconstruction performance through multi-anatomy collaborative learning with three of the proposed learners.
• Showcased the framework's ability to integrate sequence-specific learners for improved multi-pulse sequence MRI reconstruction.
• Validated the effectiveness of the

Conclusions:

• The proposed framework successfully balances shared and specific knowledge for improved MRI reconstruction.
• This approach offers a more efficient and effective alternative to traditional anatomy-specific models.
• The framework shows promise for enhancing various MRI applications, including multi-sequence reconstruction.