TITLE:

Channelized Hotelling Observer Study of Myocardial Perfusion SPECT: Effects of Quantum Noise and Patient Variation when Using OSEM Algorithm

 

AUTHORS:

X. He, J. M. Links, B. M. W. Tsui and E.C. Frey

 

PURPOSE:

Performance on a defect detection task in myocardial perfusion SPECT (MPS) is limited by both the quantum noise and the anatomic and uptake variability in patient populations. The goal of this work was to study the relative importance of these two effects and thus provide insight into the tradeoff between acquisition time, which determines the count level, and performance on a MP defect detection task.

 

METHODS AND MATERIALS:

Because human observer studies are time consuming, the Channelized Hotelling observer (CHO) is often used to estimate human performance. The CHO was applied to MPS images obtained from Monte Carlo simulated projection data of a previously-developed population of 3D NCAT-based phantoms. We generated a series of 11 sets of projection populations using the same phantom population to keep the patient variation constant. The projection populations have count levels ranging from 2-7 times clinical count level to almost noise free level to simulate different levels of quantum noise. The projection data were reconstructed using the OS-EM algorithm with attenuation, detector response and scatter compensation (ADS) and with no compensation. Post-processing to generate short-axis slices was then performed. The CHO was applied to the short-axis images and the area under the ROC curve (AUC) was computed. For each noise level and reconstruction method, we optimized the number of iterations and cutoff frequencies of the Butterworth filter to maximize the AUC.

 

RESULTS:

Both theoretical and simulation studies demonstrate that the CHO performance is dependent on the relative magnitude of the quantum noise and the patient variation, which are characterized by quantum noise covariance and patient covariance matrices, respectively. When the count level is high, the patient variation dominates, and the AUC changes very slowly with changes in the count level for the same patient variation. When the count level is low, however, the quantum noise dominates, and count level changes result in large changes in the AUC. In the simulation study we found that clinical count levels are at the transition between the two regimes. We also observed that images reconstructed with ADS compensation were better than those with no compensation for count levels as low as ½ of the standard clinical count level.

 

CONCLUSIONS:

The results of this study indicate that, at least for the CHO, at current clinical count levels there is some room to reduce acquisition time without substantially degrading performance on a MP defect detection. Also, ADS compensation can be used to allow reduced acquisition time without compromising defect detectability. Human observer studies are needed to see whether this relationship holds for human observer.

 

FUNDING SOURCES:

1 R01 HL68575