QUANTITATIVE METRICS FOR BENCHMARKING MEDICAL IMAGE HARMONIZATION.

Area of Science:

• Medical Imaging
• Image Processing
• Quantitative Analysis

Background:

• Medical image harmonization addresses domain shifts from diverse acquisition protocols.
• Benchmarking harmonization techniques is difficult due to the lack of standardized ground truth datasets.

Purpose of the Study:

• To propose novel, ground truth-free metrics for evaluating medical image harmonization.
• To assess the utility of these metrics in real-world scenarios.

Main Methods:

• Developed two intensity harmonization metrics and one anatomy preservation metric.
• Validated metrics against established image quality assessment metrics using a dataset with ground truth.
• Demonstrated application in scenarios lacking ground truth.

Main Results:

• Proposed metrics show correlation with established image quality assessment metrics.
• Metrics are applicable to real-world medical imaging harmonization tasks without ground truth.
• Provided interpretation guidelines for metric values.

Conclusions:

• Advocate for the adoption of these quantitative harmonization metrics.
• These metrics can serve as a standard for benchmarking harmonization technique performance.
• Facilitates more reliable evaluation of medical image harmonization.