Networks proving efficient enough would
- then be deployed on an open platform to be further tested by other teams.
- be subjected to analysis of the features found to be predictive of a particular event, to try to hint at the underlying pathophysiology or anatomical structures involved
- test or enhance the classification efficiency of published artificial neural networks classifiying chest CT-scans according to COVID19 diagnosis by increasing the number of available images and sources (as currently published networks are usually trained and tested on extremely limited datasets)
- transfer weights from these re-trained neural networks towards networks predicting a new outcome of interest to clinicans, but for which available data for learning is scarce, such as future need for intubation
- make these networks available on our open platform (https://imjoy.io/#/) to facilitate testing and further training by other teams
- describe which anatomical features could underly the neural network predictions
Hoa NGUYEN THI THANH