Methods for Enhancing the Robustness of the Generalized Contrast-to-Noise Ratio.

Area of Science:

• Medical Imaging
• Signal Processing
• Quantitative Analysis

Background:

• Traditional lesion detectability metrics can be manipulated by advanced techniques.
• The generalized contrast-to-noise ratio (gCNR) offers improved robustness against transformations and dynamic range alterations.
• Accurate estimation of probability density functions (PDFs) is crucial for reliable metric performance.

Purpose of the Study:

• To investigate the limitations of histogram-based gCNR implementations.
• To evaluate alternative methods for robust gCNR estimation.
• To introduce a novel approach using empirical cumulative distribution functions (eCDFs).

Main Methods:

• Simulated lesions were used to test gCNR implementation under varying data amounts and bin numbers.
• Extreme transformations were introduced to assess histogram performance.
• Parametric gCNR, robust histogram methods, and eCDF-based gCNR were evaluated.

Main Results:

• Improperly designed histograms can be poor PDF estimators, impacting gCNR accuracy.
• Histograms on rank-ordered data or with variable bin widths showed improved consistency.
• Estimating gCNR using eCDFs provided robust and consistent results.

Conclusions:

• The implementation of histogram-based gCNR requires careful consideration of binning strategies.
• Empirical cumulative distribution functions (eCDFs) offer a more reliable method for gCNR estimation.
• Rank-ordered data or variable bin widths enhance histogram-based gCNR accuracy.