An Application of Channelized Hotelling Observer for Optimization of CT Scanning Protocols

Abstract: Low contrast detection tasks like detecting a lesion in an image are of high importance in medical image diagnostics. If performed by human observers, the results are subjective, and the reproducibility is limited. This represents a problem in evaluating medical devices like computer tomography scanners during the optimization of imaging protocols; many images obtained using different settings have to be evaluated. To address the problem, the concept of model observers was developed. One of such observers is the Channelized Hotelling Observer (CHO). This work aimed to evaluate the performance of the CHO for the low contrast detectability task. Specifically, it aimed to compare the performance of two types of CT scanners: a traditional scanner (Siemens SOMATOM Force) and a prototype of the new photon-counting scanner (Siemens Count Plus). To achieve this, software that would simplify the task should be developed. A 10 channels D-DoG CHO model was applied to image sets obtained by scanning a standard QRM phantom at three dose levels and reconstructing the images using selected algorithms. The procedure was applied to the Force and Count Plus scanners. The performance of the two scanners was compared using the ROC-AUC curve and the signal-to-noise ratio (SNR). The results show that the 10-channel CHO was able to detect the low contrast detail signals for both scanners. In general, reconstruction algorithms on the Force scanner performed better than algorithms on the Count Plus scanner. The reasons for this unexpected behavior were not known. The measurements should be redone with a more suitable phantom. The newly developed software for processing and analysis was written in Python, Matlab, and R language and will be instrumental in repeating the experiment.