Comparison of channelized hotelling and human observers in determining optimum OSEM reconstruction parameters for myocardial SPECT

The performance of the Channelized Hotelling Observer (CHO) was compared to that of human observers for determining optimum OSEM reconstruction parameters for the task of defect detection in myocardial SPECT images. The OSEM reconstruction parameters varied were the number subsets/iteration and the number of iterations. The optimum parameters were those that maximized defect detectability in the SPECT images. Low noise, parallel SPECT projection data, with and without an anterior, inferior or lateral LV wall defect, were simulated using Monte Carlo code. Poisson noise was added to generate noisy realizations. Data were reconstructed using OSEM at 1&4 subsets/iteration and at 1, 3, 5, 7&9 iterations. Images were converted to 2D short-axis with integer pixel values. The CHO used 7 radially-symmetric, 2D channels, with varying levels of internal observer noise. For each parameter setting, 200 defect-present and 200 defect-absent image vectors were used to calculate the AUC. The human observers rated the likelihood that a defect was present in a specified location. For each parameter setting AUC was estimated from 48 defect-present 48 defect-absent images. The CHO results showed that the ranking of the AUC values varied with varying levels of internal noise. The averaged human observer results showed the optimum parameter setting to be 1 subset/iteration and 5-9 iterations. In our study, the CHO performance matched the human observer performance at an internal observer noise level of ∼15 used in the CHO. We conclude that addition of internal observer noise to the CHO is important. Further studies are needed to determine the dependence of the internal observer noise level on the type of images and the performance task for the CHO model to match human observer performance.