Rapid Quality Assessment of Nonrigid Image Registration Based on Supervised Learning

Area of Science:

• Medical Imaging
• Computer-Aided Diagnosis
• Interventional Radiology

Background:

• Overlaying preprocedural images onto intraprocedural images enhances visualization in interventional procedures.
• Challenges like artifacts and motion can compromise the accuracy of multimodality image registration, impacting procedural outcomes.
• Accurate registration is crucial for effective image guidance during interventions.

Purpose of the Study:

• To develop a novel framework for accurate, near real-time assessment of nonrigid multimodality image registration quality.
• To provide clinicians with a reliable method to evaluate the success of image registration during interventional procedures.

Main Methods:

• Developed a framework utilizing rapidly computable registration quality metrics.
• Combined metrics into a single binary assessment (successful or poor).
• Employed supervised learning, training and testing on clinical data with expert-generated quality metrics as ground truth.

Main Results:

• The framework achieved 81.5% accuracy in identifying successful image registration cases.
• The classification results were generated in 5.5 seconds, suitable for clinical interventional radiology.
• Demonstrated the framework's ability to provide clinicians with timely feedback on registration quality.

Conclusions:

• The proposed framework accurately assesses nonrigid multimodality image registration quality in near real-time.
• Supervised learning enables a reliable quality assessment, enhancing the safety and efficacy of image-guided interventions.
• This tool can assist clinicians by confirming or cautioning registration results during procedures.