Deep learning based COVID-19 diagnosis system

This work was supported by Korea Foundation for Women In Science, Engineering and Technology (WISET) grant funded by the Ministry of Science and ICT(MSIT) under the team research program for female engineering students.

CT scan has higher sensitivity in diagnose COVID-19 than RT-PCR. Accordingly, Deep-learing based COVID-19 diagnosis system using CT scan has been developed and used as supplementary tool for medical experts or for automatic diagnosis.

The DL networks used for existing COVID-19 diagnosis algorithms are generally black-box model with low explainablity, where the reasoning of predictions and the validity of the derivations are not probided. Therefore, these systems cannot ensure the security and reliability of its diagnosis.

Other systems that give explainability transform the models then explain their predictions, but they only offer the local explanations on specific prediction other than the actions of the whole system.

The goal of this project is to develop a COVID-19 diagnosis system with a novel architecture where derivation between cause and prediction is expressible. The system can thereby provide the principle of overall system and act as a framework for interactions between human knowledge and data-driven knowledge.