Automatic parameter selection in PET image reconstruction based on no-reference image quality assessment

In tomographic image reconstruction there is a trade-off between resolution and noise in the images. In order to achieve the best images, parameters of the reconstruction method may be varied, like the type of filter in filtered back-projection (FBP) or the number of iterations in iterative methods. The optimal value of these parameters depends on many different factors, like the amount of data acquired, and the distribution of the activity in the field-of-view. Therefore, it would be desirable to have an automatic method to select optimal reconstruction parameters. Several no-reference image quality assessment methods have been recently presented. Using an analysis of the image content or the anisotropy of the image entropy, it is possible, in principle, to select automatically the image with the best quality from a set of images reconstructed with different parameters. In this work we apply these techniques to images reconstructed from realistic simulations performed with PeneioPET for the small-animal Argus PET/CT scanner. The images selected as optimal by the method were in good agreement with the ones selected by a human observer by visual inspection. This proves that no-reference image quality (IQ) assessment methods may be able to provide a metric to select optimal reconstruction parameters.