Estimation in medical imaging without a gold standard.

Area of Science:

• Medical Imaging
• Statistical Modeling
• Biostatistics

Background:

• Physicians estimate patient parameters (e.g., cardiac ejection fraction) for diagnosis.
• Multiple estimation methods often exist, but a gold standard is rare, hindering comparison.
• Evaluating and comparing these methods is challenging without a benchmark.

Purpose of the Study:

• To examine a novel method for evaluating different estimation techniques.
• The method aims to assess estimation methods without relying on a gold standard.

Main Methods:

• The approach is analogous to regression analysis without an x-axis.
• Requires multiple estimates per patient and assumes a parameterized distribution for true values.
• Employs a statistical model linking the true parameter to its estimates to derive model and distribution parameters.

Main Results:

• Applied to simulated cardiac ejection fraction data with variations in patient numbers, modalities, and noise.
• Tested on both linear and nonlinear models, comparing performance against conventional regression.
• The method's performance trends mirrored conventional regression with varying patient numbers and noise levels.

Conclusions:

• The proposed method accurately estimates model parameters.
• These parameter estimates can effectively rank different estimation systems for specific tasks.