Image Registration Based on Autocorrelation of Local Structure

Area of Science:

• Medical imaging
• Image registration
• Computational anatomy

Background:

• Accurate medical image registration is vital for clinical applications.
• Intra-image signal fluctuations pose significant challenges to image registration accuracy.
• Existing similarity measures often struggle with space-variant intensity distortions.

Purpose of the Study:

• To develop a novel objective function for image registration that is robust to intensity variations.
• To introduce a new similarity measure based on local phase features for enhanced image registration.
• To evaluate the performance of the proposed method against established techniques.

Main Methods:

• The proposed method embeds local phase features from the monogenic signal into the Modality Independent Neighborhood Descriptor (MIND).
• Image similarity is quantified using the autocorrelation of local structure (ALOST).
• ALOST exhibits low sensitivity to intensity distortions and high distinctiveness for salient features like edges.

Main Results:

• ALOST outperformed Normalized Mutual Information (NMI) and MIND similarity measures on thoracic CT, and synthetic and real abdominal MR datasets.
• The method demonstrated robustness against space-variant intensity distortions.
• Registration of Crohn's disease MR images showed the highest correlation (r=0.56) with the endoscopic index of severity.

Conclusions:

• The ALOST method offers a robust and accurate approach for medical image registration, particularly in the presence of signal fluctuations.
• The proposed technique shows promise for improving the quantitative analysis of diseases like Crohn's using medical imaging.
• ALOST represents a significant advancement in objective functions for image similarity assessment in medical contexts.