kbressem/ct-pathology-chest-seg

CT Chest Pathology Segmentation

Deep learning framework for segmentation of chest pathologies on CT imaging, comparing U-Net, Hierarchical Dual-Resolution Network, and Prototype-Guided Segmentation Network.

Overview

Segmentation of pleural effusion, pericardial effusion, and pneumothorax using 5-fold cross-validation on 200 CT volumes.

Model	Pleural Effusion	Pericardial Effusion	Pneumothorax	Mean Dice
U-Net	0.797 ± 0.132	0.453 ± 0.216	0.452 ± 0.241	0.665 ± 0.158
Hierarchical	0.825 ± 0.117	0.535 ± 0.214	0.544 ± 0.271	0.713 ± 0.147
Prototype	0.819 ± 0.131	0.538 ± 0.229	0.567 ± 0.261	0.712 ± 0.153

Code

https://github.com/kbressem/ct-chest-pathology-seg/