Validation of a nonrigid registration error detection algorithm using clinical MRI brain data.

Area of Science:

• Medical image processing
• Computational anatomy
• Image registration quality assessment

Background:

• Nonrigid registration error identification is crucial in medical imaging.
• Previous work introduced the AQUIRC algorithm for simulated cases.
• Clinical validation of AQUIRC for local error detection is needed.

Purpose of the Study:

• To quantitatively assess AQUIRC's ability to detect local nonrigid registration errors at specific clinical landmarks.
• To validate AQUIRC's performance against established methods like LNCC.
• To explore the complementarity of AQUIRC and LNCC for improved registration error assessment.

Main Methods:

• Utilized five different registration methods with 100 target and nine atlas images.
• Quantitatively validated AQUIRC at the anterior and posterior commissure landmarks.
• Computed Local Normalized Cross-Correlation (LNCC) for comparison.
• Performed multi-linear regression combining AQUIRC and LNCC measures.

Main Results:

• AQUIRC's quality measure showed a correlation (R^2=0.542) with true target registration error (TRE) at landmarks.
• AQUIRC performed similarly to LNCC in assessing registration quality.
• Combined AQUIRC and LNCC measures demonstrated a higher correlation with TRE than either measure alone, indicating complementarity.
• The AQUIRC algorithm demonstrated potential for reducing registration errors across tested methods.

Conclusions:

• AQUIRC is a valuable tool for detecting local nonrigid registration errors in medical images.
• The combination of AQUIRC and LNCC offers a more robust assessment of registration quality.
• AQUIRC has the potential to improve the accuracy of nonrigid image registration in clinical applications.