Advancing COVID-19 Detection in 3D CT Scans

TL;DR

This work tackles COVID-19 detection from 3D chest CT scans by removing non-lung regions to concentrate on lesion-relevant voxels and reduce computation. It employs a 3D ResNeSt50 backbone initialized with COVID-19-specific pretrained weights (CMC v1) and trains via cross-entropy loss for binary classification, achieving a Macro F1 of 0.94 on the Challenge I validation set—an improvement of 16% over a baseline. The approach is validated on the COV19-CT-DB dataset, incorporating Challenge I and II data to enhance generalization, with explicit data-preprocessing and augmentation strategies. The results demonstrate that lung-focused 3D CT processing with domain-pretrained backbones can yield robust COVID-19 detection with practical efficiency.

Abstract

To make a more accurate diagnosis of COVID-19, we propose a straightforward yet effective model. Firstly, we analyse 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 ResNeSt50 as the strong feature extractor, initializing it with pretrained weights which have COVID-19-specific prior knowledge. Our model achieves a Macro F1 Score of 0.94 on the validation set of the 4th COV19D Competition Challenge $\mathrm{I}$, surpassing the baseline by 16%. This indicates its effectiveness in distinguishing between COVID-19 and non-COVID-19 cases, making it a robust method for COVID-19 detection.

Figures (2)

• Figure 1: Samples of non-COVID-19 and COVID-19 from the COV19-CT-DB database.
• Figure 2: Overview of our framework for COVID-19 detection.