In most recent computer aided celiac disease diagnosis approaches, image regions (patches) showing discriminative features necessarily need to be manually extracted by the medical doctor, prior to the automated classification pipeline. However, although the obtained classification outcomes based on such semi-automated systems are attractive, a human interaction finally is undesired. In this work, fully automated approaches are investigated which are based on the measurement of several image quality properties. Firstly, we investigate a method based on optimization of single quality measures as well as an approach based on weighted combinations of these metrics. Furthermore, a weighted decision-level and a weighted feature-level fusion method are investigated which are not based on the selection of one single best patch, but on a weighted combination. In a large experimental setting, we evaluate these methods with respect to the achieved overall classification rates. Finally, especially the proposed feature-level fusion method supplies the best performances and comes close to manual experts' patch selection as far as the accuracy is concerned.