MultiMorph: On-demand Atlas Construction.

Area of Science:

• Neuroimaging
• Computational Biology
• Medical Image Analysis

Background:

• Anatomical atlases are crucial for understanding brain structure and variability.
• Current atlas construction methods are computationally intensive, hindering rapid research.
• Reliance on mismatched atlases negatively impacts downstream scientific analyses.

Purpose of the Study:

• To introduce MultiMorph, a novel, fast, and efficient method for on-the-fly anatomical atlas construction.
• To enable rapid generation of high-quality, population-specific atlases for 3D brain datasets.
• To overcome the computational limitations of existing atlas construction techniques.

Main Methods:

• MultiMorph utilizes a feedforward model for single-pass atlas generation.
• Employs a linear group-interaction layer for feature aggregation and sharing.
• Leverages auxiliary synthetic data for generalization to new imaging modalities and populations.

Main Results:

• MultiMorph achieves a 100-fold reduction in computation time compared to state-of-the-art methods.
• Demonstrates superior performance in both small and large population settings.
• Produces high-quality, population-specific atlases without fine-tuning or optimization.

Conclusions:

• MultiMorph provides an accessible framework for rapid, high-quality anatomical atlas generation.
• Facilitates diverse biomedical research by enabling quick, population-specific atlas creation.
• Empowers researchers without machine learning expertise to perform advanced neuroimaging analyses.