Multiatlas-based segmentation with preregistration atlas selection.

Area of Science:

• Medical Image Analysis
• Computational Anatomy

Background:

• Multi-atlas segmentation enhances robustness in medical image analysis.
• High computation time is a major drawback due to extensive atlas registration.

Purpose of the Study:

• To reduce the computational load of multi-atlas segmentation.
• To achieve this by implementing a heuristic atlas selection strategy before registration.

Main Methods:

• Pairwise registrations are performed for atlas clustering in a preprocessing step.
• Cluster representatives are registered to the target image, with quality estimation guiding further registration.
• Segmentations are fused post-registration to create a final output.

Main Results:

• The proposed preregistration method was compared against post-registration selection using 182 prostate cancer patient atlases.
• A slight, statistically insignificant difference in accuracy was observed between preregistration and post-registration selection.
• The method achieved an average reduction of 80% in the number of atlases requiring registration.

Conclusions:

• The heuristic preregistration atlas selection effectively reduces computational burden.
• This method maintains segmentation accuracy comparable to post-registration selection.
• It offers a computationally efficient alternative for multi-atlas segmentation workflows.