| Zugriffsnummer | 40192 |
| Dokumenttyp | Zeitschriftenartikel |
| Peer Review | mit Peer Review |
| Sprache | Englisch |
| Titel | A simple parametric model observer for quality assurance in computer tomography |
| Autor(in); Institution |
Anton, Mathias; 6.2, Dosimetrie für Strahlentherapie und Röntgendiagnostik, PTB-Braunschweig
Khanin, Alexander; 6.2, Dosimetrie für Strahlentherapie und Röntgendiagnostik, PTB-Braunschweig
Kretz, Tobias; 8.4, Mathematische Modellierung und Datenanalyse, PTB-Berlin
Reginatto, Marcel; 6.4, Neutronenstrahlung, PTB-Braunschweig
Elster, Clemens; 8.4, Mathematische Modellierung und Datenanalyse, PTB-Berlin
|
| Quelle/Jahr | Physics in Medicine and Biology: 63 (2018), 7, 1 - 16 |
| Artikelnummer | 075011 |
| ISSN | 1361-6560 (ONLINE) ; 0031-9155 (PRINT) |
| DOI | |
| Verlag | Bristol: IOP Publishing |
| Freie Schlagworte | image quality ; task specific quality assessment ; Quality assurance ; Bayes ; AUC ; CT ; x-ray tomography ; model observer |
| Zusammenfassung | Model observers are mathematical classifiers that are used for the quality assessment of imaging systems such as computer tomography. The quality of the imaging system is quantified by means of the performance of a selected model observer. For binary classification tasks, the performance of the model observer is defined by the area under its ROC curve (AUC). Typically, the AUC is estimated by applying the model observer to a large set of training and test data. However, the recording of these large data sets is not always practical for routine quality assurance. In this paper we propose as an alternative a parametric model observer that is based on a simple phantom, and we provide a Bayesian estimation of its AUC. It is shown that a limited number of repeatedly recorded images (10-15) is already sufficient to obtain results suitable for the quality assessment of an imaging system. A MATLAB® function is provided for the calculation of the results. The performance of the proposed model observer is compared to that of the established channelised Hotelling observer (CHO) and the nonprewhitening matched filter (NPW) for simulated images as well as for images obtained from a low-contrast phantom on an x-ray tomography scanner. The results suggest that the proposed parametric model observer, along with its Bayesian treatment, can provide an efficient, practical alternative for the quality assessment of CT imaging systems. |
