Retrospective intermodality registration techniques for images of the head: surface-based versus volume-based

Area of Science:

• Medical Imaging
• Image Registration
• Neurosurgery

Background:

• Image registration aligns medical scans from different modalities (CT, MR, PET).
• Previous evaluations assessed techniques individually.
• Grouping retrospective techniques by volume-based or surface-based methods offers new insights.

Purpose of the Study:

• To blindly evaluate and compare two groups of retrospective image registration techniques.
• To establish a prospective marker-based registration method as the gold standard.
• To analyze performance differences between volume-based and surface-based registration groups.

Main Methods:

• Retrospective registrations were performed blindly by external collaborators.
• Image volumes from CT, MR, and PET scans of neurosurgery patients were used.
• Accuracy was evaluated using multiple volumes of interest (VOIs) against a gold standard.

Main Results:

• Volume-based techniques were substantially more accurate and reliable for CT-to-MR registration.
• Volume-based techniques showed slightly better accuracy for PET-to-MR registration.
• Surface-based techniques exhibited a more pronounced rotational error component.

Conclusions:

• Volume-based image registration generally outperforms surface-based methods in neurosurgical applications.
• All examined techniques can yield satisfactory results, but visual inspection is vital to detect substantial errors.
• Grouping registration techniques reveals performance trends not evident in individual assessments.