RS-MOCO: A deep learning-based topology-preserving image registration method for cardiac T1 mapping

Area of Science:

• Medical Imaging
• Artificial Intelligence

Background:

• Cardiac T1 mapping is crucial for assessing myocardial tissue but lacks effective motion correction.
• Existing methods struggle with robustness and efficiency in cardiac T1 mapping.

Purpose of the Study:

• To develop a deep learning-based, topology-preserving image registration framework for robust motion correction in cardiac T1 mapping.
• To enhance the accuracy and efficiency of motion correction in cardiac T1 mapping.

Main Methods:

• Proposed a deep learning framework integrating a novel implicit consistency constraint (BLOC) for topology preservation.
• Introduced a weighted image similarity metric to handle contrast variations in multimodal registration.
• Incorporated a semi-supervised myocardium segmentation network and a dual-domain attention module.

Main Results:

• The framework demonstrated high effectiveness and robustness in motion correction for cardiac T1 mapping.
• The weighted image similarity metric significantly improved motion correction efficacy.
• The BLOC constraint ensured desirable topology-preserving registration mapping, validated by comparative and ablation studies.

Conclusions:

• The proposed deep learning framework offers an effective, robust, and efficient solution for motion correction in cardiac T1 mapping.
• The novel BLOC constraint and weighted similarity metric are key innovations enhancing registration accuracy and topology preservation.