Core Concepts
Pseudo labels enhance COVID-19 detection accuracy through domain adaptation, addressing data scarcity challenges in healthcare crises.
Abstract
Standalone Note here
Abstract
Two-stage framework using pseudo labels for domain adaptation in COVID-19 detection from CT scans.
Model improves accuracy and adaptability across medical centers.
Introduction
Efficient diagnostic methods crucial for managing COVID-19 outbreak.
CT imaging valuable but interpretation challenging.
Methodology
Framework structured into two stages: training on annotated data and generating pseudo labels for non-annotated data.
Datasets
COV19-CT-DB database contains 7,756 3D chest CT scans.
Training set includes 239 annotated 3D CT scans.
Experiments
Data pre-processing involves resizing and intensity normalization of CT volumes.
Implementation details include the use of ResNest50 as backbones and Adam optimization algorithm.
Conclusion
Proposed framework enhances diagnostic capabilities by leveraging pseudo labels for domain adaptation.
References
Stats
COV19-CT-DBには、7,756の3D胸部CTスキャンが含まれています。
訓練セットには、239の注釈付き3D CTスキャンが含まれています。