Comparing endoscopic imaging configurations in computer-aided celiac disease diagnosis

During the past years, significant research has been done on computer aided celiac disease diagnosis based on endoscopic image data. Mostly the aim is to increase the final classification accuracies based on a certain endoscopic data set. In this work, we investigate the impact of different endoscopic devices as well as imaging modalities with distinct impact on the visual properties of the image data. Apart from the obtained classification accuracies, special focus is on a metric measuring the relative performance of computer based methods compared to the visual classification performance of human experts on the respective data sets. Thereby a potential bias due to variations between the data sets can be circumvented. Finally, we can make statements on the performances of computer based methods compared to human experts considering the same data set, and, based on previous knowledge, assumptions on the most appropriate imaging setting in case of computer aided diagnosis can be made.