Validation of DRAMMS among 12 Popular Methods in Cross-Subject Cardiac MRI Registration

Area of Science:

• Medical Imaging
• Computational Anatomy
• Cardiovascular Research

Background:

• Cross-subject image registration is crucial for cardiac studies.
• Voxel-wise methods are common but their accuracy and stability lack comparative studies.
• Evaluating existing and novel methods is needed for robust cardiac image analysis.

Purpose of the Study:

• To evaluate and compare the accuracy and stability of 12 popular image registration methods for cross-subject cardiac applications.
• To validate the performance of the DRAMMS registration method in this specific context.
• To correlate registration accuracy with deformation characteristics.

Main Methods:

• A dataset of 552 short-axis end-diastole cardiac MR image pairs from 24 subjects was used.
• 12 registration methods, including DRAMMS, ANTs, MI-FFD, Demons, and ART, were applied.
• Accuracy was assessed using Jaccard overlap between deformed and target expert annotations.
• Deformation aggressiveness was measured using Jacobian determinants.

Main Results:

• DRAMMS achieved high accuracy and balanced accuracy with deformation aggressiveness.
• Following DRAMMS, ANTs, MI-FFD, Demons, and ART showed notable performance.
• Findings align with previous research on cross-subject brain image registration.

Conclusions:

• DRAMMS is a highly accurate and stable method for cross-subject cardiac image registration.
• The study provides a comparative analysis to guide method selection in cardiac imaging research.
• Results emphasize the importance of evaluating registration methods for specific anatomical regions and study types.