The paper titled "COVID detection and severity prediction with 3D-ConvNeXt and custom pretrainings" by Daniel Kienzle, Julian Lorenz, Robin Schön, Katja Ludwig and Rainer Lienhart is accepted at the Workshop on AI-enabled Medical Image Analysis at ECCV 2022.

In this paper, the authors present how the ConvNeXt architecture can be leveraged for the classification of 3D-CT scans. Particularly, various transfer learning methods supporting the application on 3D medical data are explored. With the insights presented in this paper, the authors achieve the 2nd place in the 1st COVID19 Severity
Detection Challenge and the 3rd place in the 2nd COVID19 Detection Challenge.

Since COVID strongly affects the respiratory system, lung CT-scans can be used for the analysis of a patients health. We intro duce a neural network for the prediction of the severity of lung damage and the detection of a COVID-infection using three-dimensional CT-data. Therefore, we adapt the recent ConvNeXt model to process three-dimensional data. Furthermore, we design and analyze different pretraining methods specifically designed to improve the models ability to handle three-dimensional CT-data. We rank 2nd in the 1st COVID19 Severity Detection Challenge and 3rd in the 2nd COVID19 Detection Challenge.