The main theme will be to model the total diagnostic image processing chain, so that weak or critical points in algorithms can be identified and improved already in early stages of research. Specifically the models will allow estimating the robustness of a combination of a cascade of image processing and analysis operaters w.r.t. image nois, suboptimal parameter setting etc. To ensure that the methodology will eventually be applicable in practice, we will study multimodal video-based lung cancer diagnosis as an important test case.