Numerical stability issues on channelized Hotelling observer under different background assumptions.

Area of Science:

• Medical Imaging
• Image Quality Assessment
• Observer Performance Modeling

Background:

• The channelized Hotelling observer (CHO) is widely used for medical image quality assessment.
• A known issue is that CHO detection performance does not increase with the number of channels, contrary to theoretical expectations.
• The underlying reason for this limitation has not been previously identified in the field.

Purpose of the Study:

• To identify the cause of the numerical instability in the channelized Hotelling observer (CHO).
• To propose a method for resolving the identified instability issue.
• To enable reliable detection performance calculations for channelized model observers in medical imaging.

Main Methods:

• Analysis of the channelized Hotelling observer (CHO) formulation.
• Identification of the ill-posed nature of the scatter matrix as the root cause of instability.
• Application of Tikhonov regularization to stabilize the scatter matrix calculation.

Main Results:

• Demonstrated that the ill-posed scatter matrix is responsible for the non-increasing detection performance of the CHO with more channels.
• Successfully applied Tikhonov regularization to address the numerical instability.
• Showcased a novel application of Tikhonov regularization within the domain of medical image analysis.

Conclusions:

• The numerical instability in CHO performance is attributed to the ill-posed scatter matrix.
• Tikhonov regularization provides an effective solution, ensuring reliable CHO performance assessment.
• This work is crucial for advancing research in channelized model observers and medical image quality evaluation.