Use of Sub-Ensembles and Multi-Template Observers to Evaluate Detection Task Performance for Data That are Not...

Area of Science:

• Medical Imaging
• Image Quality Assessment
• Observer Performance

Background:

• The Hotelling Observer (HO) is a standard for medical image quality evaluation.
• Standard HO application is suboptimal for non-multivariate-normal (non-MVN) data.
• Developing robust methods for non-MVN data is crucial for accurate image quality assessment.

Purpose of the Study:

• To introduce and evaluate two multi-template linear observer strategies for non-MVN data.
• To compare these strategies against the single-template HO for optimizing reconstruction parameters.
• To determine the most effective strategy for enhancing image quality assessment in challenging datasets.

Main Methods:

• Dividing non-MVN data ensembles into multivariate-normal (MVN) and homoscedastic sub-ensembles.
• Applying a modified Hotelling Observer (HO) to each sub-ensemble and averaging the area under the receiver operating characteristics curve (AUC).
• Applying a Linear Discriminant (LD) observer to estimate test statistics for each sub-ensemble and calculating a global AUC from pooled statistics.

Main Results:

• The multi-template Linear Discriminant (LD) strategy achieved the highest AUC when only shifting of HO test statistics was permitted.
• Application to myocardial perfusion SPECT studies demonstrated that the multi-template LD strategy yielded the highest AUC for given reconstruction parameters.
• Both multi-template strategies produced comparable optimal reconstruction parameters, outperforming the single-template HO strategy.

Conclusions:

• Multi-template linear observer strategies effectively handle non-MVN data in medical imaging.
• The multi-template LD strategy provides superior performance for image quality assessment and parameter optimization.
• These advanced methods enhance the reliability of image quality evaluation in complex medical imaging scenarios.