Deep Learning Frameworks for COVID-19 CT Lesion Segmentation Compared
A new study evaluates deep learning architectures for segmenting COVID-19 lesions in CT images. Researchers integrated four architectures—Unet, PSPNet, Linknet, and FPN—with six pre-trained encoders: VGG 19, DenseNet 121, Inception ResNet V2, MobileNet V2, and SeresNe. The work aims to address the lack of standardized methodology in medical image segmentation by providing a comprehensive performance benchmark. The study's findings could serve as a reference for segmentation in other imaging scenarios.
Key facts
- Study compares four deep learning architectures: Unet, PSPNet, Linknet, FPN
- Six pre-trained encoders used: VGG 19, DenseNet 121, Inception ResNet V2, MobileNet V2, SeresNe
- Focus is on predicting COVID-19 lesions in CT images
- Research aims to standardize performance analysis in medical image segmentation
- Results may apply to other imaging scenarios
- Published as arXiv preprint 2605.20459
Entities
Institutions
- arXiv