Główne pojęcia
Proposing COVID-CT-H-UNet for improved COVID-19 CT segmentation.
Streszczenie
The paper introduces COVID-CT-H-UNet, a novel network for COVID-19 CT segmentation. It addresses challenges faced by existing methods, such as misclassification of normal pixels and hazy borders. By combining an attention mechanism and Bi-category Hybrid Loss, the proposed network enhances segmentation accuracy. Experimental results show significant improvement over previous models in identifying clinical COVID-19 from CT images.
Statystyki
"Around 560 million cases were confirmed and approximately 6 million patients died."
"Only 20 CT scans make up this dataset for segmentation."
"Trained the network for around 100 epochs with a batch size of 32."
Cytaty
"Since COVID-19 currently has no viable treatments, early detection becomes crucial."
"The proposed COVID-CT-H-UNet’s segmentation impact has greatly improved."
"Our proposed model outperforms all of them by a significant margin in Dice and Specificity metrics."