Automatic quantification of multi-modal rigid registration accuracy using feature detectors.

Area of Science:

• Medical Imaging
• Radiotherapy
• Image Registration

Background:

• Multi-modal imaging enhances tumor delineation in radiotherapy.
• Accurate 3D/3D registration is crucial for aligning images from different modalities.
• Current validation methods (visual inspection, fiducials) have limitations.

Purpose of the Study:

• To develop and evaluate an automatic, non-invasive method for assessing multi-modal rigid image registration quality.
• To compare feature detectors (SURF, Harris) against manual and fiducial-based evaluations.

Main Methods:

• Implemented an automatic registration quality assessment using Speeded-Up Robust Features (SURF) and Harris feature detectors.
• Calculated registration quality based on the mean Euclidean distance between matching interest point pairs.
• Validated the method on ex vivo porcine skull and in vivo brain and lung datasets using qualitative and quantitative measures.

Main Results:

• The Harris detector demonstrated better performance than SURF, yielding registration error estimates closer to the gold standard.
• In lung cases, the Harris detector's mean target registration error (mTRE) was within 1 mm of manual annotation.
• Both automatic methods showed comparable results to manual detection in the porcine skull dataset, though both overestimated fiducial-based mTRE.

Conclusions:

• The Harris feature detector is a suitable method for automatically quantifying the geometric accuracy of multi-modal rigid image registration.
• This automated approach offers a more objective and potentially more clinically accepted alternative to current validation techniques.