No-gold-standard evaluation of image-acquisition methods using patient data

Area of Science:

• Medical Imaging
• Quantitative Imaging Analysis
• Biomedical Engineering

Background:

• Developing reliable quantitative imaging requires objective evaluation of new image-acquisition methods (IAMs).
• Clinical evaluation typically needs a gold standard, which is often unavailable with patient data.
• Existing no-gold-standard (NGS) techniques require extensive patient scanning across all IAMs, posing practical challenges.

Purpose of the Study:

• To develop a practical NGS technique for comparing quantitative imaging methods using patient data.
• To enable evaluation when different sets of patients are scanned with different IAMs.
• To provide a method for assessing accuracy, precision, and reliability of IAMs.

Main Methods:

• Developed an NGS technique assuming a linear relationship between true and measured quantitative values (slope, bias, noise SD).
• Employed a maximum-likelihood procedure to estimate linear relationship parameters under a unimodal distribution assumption for true values.
• Utilized patient data where distinct patient sets were imaged using different IAMs.

Main Results:

• The proposed NGS technique effectively estimates IAM performance parameters from patient data with varied scanning protocols.
• Figures of merit for accuracy, precision, and reliability can be derived from the estimated parameters.
• The method allows for comparative analysis of IAMs without requiring every patient to undergo all imaging methods.

Conclusions:

• The developed NGS technique offers a clinically practical approach to evaluating and comparing quantitative imaging methods.
• Potential applications include optimization of imaging protocols, quantifying system performance differences, and system harmonization.
• This method facilitates robust IAM assessment using real-world patient data, addressing limitations of existing techniques.