Automated Chest CT Image Segmentation of COVID-19 Lung Infection

20 CT scans from COVID-19 patients were used to develop a segmentation model for lungs and infected regions. The model, based on a 3D U-Net architecture, achieved Dice similarity coefficients of 0.956 for lungs and 0.761 for infection. This dataset and model were created by Dominik Müller of the University of Augsburg to address limited public COVID-19 imaging data.

Use Cases

• Training segmentation models for COVID-19 lung infection based on annotated CT scans.
• Benchmarking new segmentation algorithms against a model achieving a 0.761 Dice score for infection.
• Developing clinical decision support tools for quantitative disease monitoring mentioned in the description.
• Studying methods to reduce overfitting in medical imaging with limited data, as described in the pipeline.

Strengths

• Model performance is quantified with Dice scores of 0.956 for lungs and 0.761 for infection.
• The method was validated using a 5-fold cross-validation on 20 CT scans.
• The pipeline specifically addresses overfitting risks common with small datasets through extensive data augmentation.

Limitations

• The dataset is described as limited to 20 CT scans, which is a small sample size.
• Row count and specific file formats for the underlying images are unknown.
• Column-level documentation for any associated metadata is absent.

Provenance

Source

University of Augsburg

Collection Method

CT scans from COVID-19 patients, processed with an automated segmentation pipeline.