Advancing COVID-19 Detection in 3D CT Scans

Qingqiu Li, Runtian Yuan, Junlin Hou, Jilan Xu, Yuejie Zhang, Rui Feng, Hao Chen; Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2024, pp. 5149-5156

To make a more accurate diagnosis of COVID-19 we propose a straightforward yet effective model. Firstly we analyze the characteristics of 3D CT scans and remove the non-lung parts facilitating the model to focus on lesion related areas and reducing computational cost. We use ResNeSt-50 as the strong feature extractor exploring various pre-trained weights and fine-tuning methods. After a thorough comparison we initialize our model with CMC v1 pre-trained weights which incorporate COVID-19-specific prior knowledge and perform Visual Prompt Tuning to reduce the number of training parameters. The superiority of our model is demonstrated through extensive experiments showing significant improvements in COVID-19 detection performance compared to the baseline model. Among 12 participating teams our method ranked 4th in the 4th COVID-19 Competition Challenge I with an average Macro F1 Score of 94.24%.