At the onset of the covid-19 pandemic in Europe, we recognised that our experience deploying AI for chest CT scans should serve the medical staff fighting the new disease. With EU support (through a Horizon 2020 grant), we set up the International Consortium for Covid-19 Imaging AI: ICOVAI, also known as the CoronAI project. Our aim was to build an AI clinical application for covid-19 imaging.
Together with academic organisations, clinicians, researchers, and industry experts, we took a shortcut-free approach to AI model development, as explained in this article. The initiative was purpose-driven and not-for-profit.
During the CoronAI project, we have taken the following steps:
- Collecting over 3,500 chest CT scans;
- Enhancing the annotator for algorithm development and completing 1,000 high-quality annotations;
- Building three AI models: lung mask; quantification of the damaged lung; and classification of COVID appearance.
The Netherlands Cancer Institute (NKI) has recently evaluated the models’ performance on an independent dataset. On July 7th, we will present the outcome of this study and its research value to consortium partners.
Based on the results and international scientific literature on the current clinical relevance of AI for covid-19 (for example, this study in Nature Medicine), we have decided not to productise the AI models into a medical device.
Nonetheless, the solid datasets, annotations, and methods we used constitute an important contribution to research. We hope a study detailing the results will be published in an academic journal later this year. In the future, we or other consortium partners may further leverage the findings from this project.
