Ching Lam Choi - Corona-Net

CoronaNet, a deep learning model for COVID-19 diagnosis confirmation using CT scans, leverages an encoder-decoder architecture and performs binary classification, binary segmentation, and three-class segmentation.

Key takeaways
  • CoronaNet is a deep learning model for COVID-19 diagnosis confirmation using CT scans.
  • It utilizes an encoder-decoder architecture with efficient net as the backbone.
  • CoronaNet performs three tasks: binary classification, binary segmentation, and three-class segmentation.
  • It leverages techniques such as scale shift and elastic deformations for data augmentation.
  • PyTorch is used for model implementation due to its research support, customization ability, and dynamic computation graphs.
  • CoronaNet achieves good results with limited data, outperforming baselines with minimal parameters.
  • Further improvements can be made by including dilation, attention mechanisms, and class activation mapping.
  • The code is open-source and available on GitHub.
  • The project aims to help with the pandemic through technology and education.