Multiscale joint segmentation and registration of image morphology

Area of Science:

• Medical Image Analysis
• Computational Imaging
• Computer Vision

Background:

• Image registration, denoising, and segmentation are interdependent tasks crucial for medical image analysis.
• Combining information from multiple image modalities improves segmentation robustness.
• Existing methods often treat these fundamental image processing tasks separately.

Purpose of the Study:

• To develop a unified variational approach for multimodal image registration.
• To integrate edge detection, edge-preserving denoising, and non-rigid morphological registration.
• To enhance the robustness and accuracy of image registration, particularly for medical applications.

Main Methods:

• A variational approach combining Mumford-Shah type free discontinuity problems for singular morphology (edge sets) and matching.
• Quantification of regular morphology matching using deformed and original normals.
• Non-linear elastic energy for controlling deformation smoothness and injectivity.
• A multi-scale approach based on phase field approximation for an efficient algorithm.

Main Results:

• The presented variational method effectively integrates denoising, edge detection, and morphological registration.
• The approach demonstrates robustness in handling structural correspondences between image modalities.
• Numerical experiments confirm the efficacy and efficiency of the multi-scale algorithm.
• Successful applications shown on medical image datasets.

Conclusions:

• The unified variational framework offers a robust solution for multimodal image registration.
• The integration of fundamental image processing tasks leads to improved segmentation and registration accuracy.
• The developed multi-scale, phase-field-based algorithm is efficient and applicable to medical imaging.