Karl Stierstorfer PhD, Presenter: Employee, Siemens AG

Frédéric Noo, Abstract Co-Author: Nothing to Disclose

Rainer Raupach PhD, Abstract Co-Author: Employee, Siemens AG

Bernhard Schmidt PhD, Abstract Co-Author: Employee, Siemens AG

Friederike Schöck, Abstract Co-Author: Employee, Siemens AG

Thomas G. Flohr PhD, Abstract Co-Author: Employee, Siemens AG

Various powerful nonlinear noise reduction techniques have been proposed for Computed Tomography, claiming substantial dose reduction potentials (e.g. Thibault et al., Med. Phys. 34, pp. 4526 (2007), Bruder et al., Adaptive Iterative Reconstruction, SPIE 2011, to be published). To prove that these dose reductions are really achievable in practice remains a difficult task, however. Meanwhile the community has realized that simply measuring the noise and sharpness of images is not sufficient. Unlike in nuclear medicine where model observers are a well-established technique used to assess e.g. various collimator designs, there is no such standard for CT. Moreover, there are serious doubts if the standard techniques used in nuclear medicine, e.g. SKE-BKV with lumpy background and circularly symmetric channels, can be applied in CT where the noise is much more correlated and directional (A. Wunderlich, F. Noo, Evaluation of the Impact of Tube Current Modulation on Lesion Detectability using Model Observers, 30th IEEE EMBS Conference, 2008).

We compare simulated images processed with and without a nonlinear adaptive raw data filter, applying standard (Channelized Hotelling) model observers to different tasks, performing ROC and LROC analysis.

We find that the choice of the observers and the task significantly influences the outcome of comparisons. We also find that the task must be carefully selected dependent on the question to be answered.

Task based image quality assessments represent a powerful tool to evaluate and optimize new non-linear image reconstruction techniques. This tool, however, comes with many degrees of freedom that will require careful analysis before they can be used as standardized black-box methods.

Nonlinear data or image processing techniques can be evaluated with model observers which have to be carefully selected to provide meaningful results.

Stierstorfer, K, Noo, F, Raupach, R, Schmidt, B, Schöck, F, Flohr, T, Using Model Observers to Evaluate the Efficiency of Nonlinear Data Processing Techniques in CT.  Radiological Society of North America 2011 Scientific Assembly and Annual Meeting, November 26 - December 2, 2011 ,Chicago IL. http://archive.rsna.org/2011/11010884.html