Nested hierarchical group-wise registration with a graph-based subgrouping strategy for efficient template...

Area of Science:

• Medical Imaging
• Computational Anatomy
• Biomedical Engineering

Background:

• Group-wise registration is crucial for creating population-level medical image templates.
• Existing methods struggle with efficiency, capacity, and large population variations.

Purpose of the Study:

• To address limitations in current group-wise registration algorithms.
• To introduce a novel Nested Hierarchical Group-wise Registration (NHGR) framework.

Main Methods:

• A subgrouping strategy divides large populations into smaller, manageable subgroups.
• Hierarchical registration is performed across subgroups at multiple scales.
• A nested approach optimizes registration at population, subgroup, and image-pair levels using a deep multi-resolution network.

Main Results:

• The NHGR framework demonstrated superior registration efficiency compared to existing methods.
• The method resulted in sharper and more central template images.
• Evaluated on adult and adolescent brain datasets, NHGR showed consistent performance improvements.

Conclusions:

• The proposed NHGR framework effectively overcomes challenges in large-scale group-wise medical image registration.
• NHGR offers a robust solution for constructing high-quality population-specific templates.
• This advancement has significant implications for population-level medical data analysis.