Performance Assessment for Brain MR Imaging Registration Methods

Area of Science:

• Medical Imaging
• Neuroimaging
• Image Registration

Background:

• Commercial brain MR imaging registration algorithms lack standardized performance metrics.
• Objective evaluation is crucial for selecting reliable registration tools.

Purpose of the Study:

• To propose a rational methodology for comparing the performance of clinical brain MR imaging registration algorithms.
• To introduce quantifiable metrics for assessing registration accuracy.

Main Methods:

• Utilized 1175 fiducial landmarks across 4 MR sequences (T2, FLAIR, SWAN, T1 postcontrast) from 20 patients.
• Applied multiple registration algorithms using T2 as a reference, calculating Euclidean error pre- and post-registration.
• Introduced Euclidean and statistical effectiveness ratios to quantify registration performance against a criterion standard.

Main Results:

• Initial registration errors varied by sequence (e.g., FLAIR: 2.07 ± 0.55 mm).
• Post-registration, significant error reductions were observed (e.g., T1 postcontrast best error: 1.06 ± 0.16 mm).
• Effectiveness ratios demonstrated varying performance across algorithms, with a commercial GE registration showing a statistical effectiveness ratio of 0.929.

Conclusions:

• The study demonstrates a robust method for comparing brain MR imaging registration algorithm performance.
• Recommends Euclidean error, Euclidean effectiveness ratio, and statistical effectiveness ratio as key performance metrics.
• These metrics facilitate rational and objective comparisons of clinical registration algorithms.