- Handbook on the Physics of Diagnostic Radiology, 2014 (IAEA)
- Software for ROC Analysis
- Dorfman D.D., Berbaum K.S., Metz, C. E., Receiver Operating Characteristic Rating Analysis: Generalization to the Population of Readers and Patients with the Jackknife Method, Invest Radiol. 27 (1992) 723
- Pan X., Metz C.E., The "proper" binormal model: parametric ROC curve estimation with degenerate data. Acad Radiol 4 (1997) 380
- Hills S.L, Obuchowski N.A., Schartz K.M., Berbaum K.S., Comparison of ROC Methods. Stat Med 24 (2005) 1579
- Schueler B.A., Gray J.E., Gisvold J.J., A comparison of mammography screen-film combinations, Features Analysis in Mammography Radiology 184 (1992)b629
For additional references click here
Assessment of image quality of clinical, patient images
The most important evaluation of image quality occurs in the clinical setting using images of clinical patients. In most instances it is not appropriate to take additional radiographs of a patient for the purpose of evaluating a different screen-film system, for example. This would result in additional radiation to the patient while providing no direct benefit to the patient in terms of clinical diagnosis. Consequently, one must consider other approaches using clinically acquired images.
One approach for comparing two similar imaging systems is to use a right-left comparison. For example, in mammography one may image the left breast on one type of imaging system and the right breast on the comparison system. This way the patient does not incur additional radiation dose over and above what would be required for a clinical study. However, there is still a potential ethical question— Is the comparison system providing adequate image quality for clinical purposes? If this is not known, then approval by an ethical review board and informed consent may be required.
Whenever one uses clinical images there is a question as to whether approval is required by an ethical panel, sometimes referred to as an institutional review board or ethical review board. In many countries such approval is needed even if only the review of patient records is required for the research.
A radiologist must be involved in any study using clinical images. This is necessary since the images belong to the radiology department and since the radiologist is the best individual to assure the study is asking relevant questions.
It is essential to start statistical analysis during the design of the study. Statistics will be used to predict the number of images and readers required to obtain statistically significant results. Consequently, it is helpful to enlist the assistance of a statistician early in the project.
Introduction to References
Metz has been a prolific researcher and author regarding receiver operating characteristic (ROC) analysis. The paper by Dorfman, Berbaum, and Metz should be the standard reference for medical physicists on ROC analysis. Metz also provides software for ROC analysis on his website.
Schueler, et al., provide an example of the application of features analysis to different screen-film mammography systems.