Weakly Supervised Learning of Metric Aggregations for Deformable Image Registration

Area of Science:

• Biomedical image computing
• Medical image analysis
• Computer-assisted surgery

Background:

• Deformable registration is crucial for aligning medical images.
• Conventional methods rely on similarity criteria, deformation models, and smoothness constraints.
• The choice of metric function significantly impacts registration accuracy.

Purpose of the Study:

• To propose a novel weakly supervised approach for learning domain-specific metric aggregations using anatomical segmentations.
• To improve the accuracy of deformable image registration by incorporating semantic information.
• To develop a multi-metric registration algorithm with a spatially varying similarity metric.

Main Methods:

• A weakly supervised approach using latent structured support vector machines to learn metric aggregations.
• Integration of learned matching criteria within a metric-free optimization framework based on graphical models.
• Development of a multi-metric algorithm with a spatially varying similarity metric conditioned on anatomical structures.

Main Results:

• Extensive evaluation on three diverse CT and MRI image datasets.
• Demonstration that the learned multi-metric registration significantly outperforms conventional single-metric approaches.
• Validation of improved accuracy through the incorporation of semantic information.

Conclusions:

• Incorporating semantic information via anatomical segmentations enhances deformable image registration accuracy.
• The proposed learned multi-metric approach offers a significant advancement over traditional single-metric methods.
• This method provides a more robust and accurate solution for medical image alignment.