COVID X-RAy
AI for more personal, powerful healthcare
Covid X-Ray
Success story
Open access scripts on model explainability for the healthcare field, as applied to Covid-19 X-rays
Explainability methods help clinicians understand how a model works. They are an additional tool that can help determine how relevant a model is to clinical applications.
Ultimately, these tools could be used in decision-assistance applications in the halthcare field.
Challenge
In order to help control the spread of Covid-19, Probayes has made its scripts on model explainability (model comparison, importance, and limitations) freely available for healthcare applications and, most notably, for Covid-19 X-rays.
The project
Compare several explainability methods (a particularly important issue in the healthcare field).
Datasets used: COVIDx3
ResNet50 pretrained on ImageNet was selected as the reference model for explainability:
– “normal” X-ray
– “pneumonia” X-ray
– “Covid-19” X-ray
Results
Studying the explainability of the model provides an opportunity to see the forest and not only the trees by stepping back from the metrics used to evaluate the results of the model and even identify areas for improvement in the model. Explainability methods allow users to validate that the areas that contribute to prediction are precisely those areas where information is located.
Case correct prediction class “normal”
Consistent positioning over the lungs
Case correct prediction class “pneumonia”
Inflammation that appears to be correctly located
Correct prediction of “Covid-19” class
Inflammation that appears to be correctly located
“Lorem ipsum dolor sit amet, consectetur adipiscing elit. Fusce eu lacus diam. Cras congue, neque ac mattis efficitur, nibh ipsum fringilla nunc, sed iaculis neque odio non orci. Vestibulum ante ipsum primis in faucibus orci luctus et ultrices posuere cubilia Curae; Praesent lacinia, urna quis rhoncus lobortis, urna neque tempus tellus, quis auctor justo lectus vitae libero. Nulla non porta odio. Donec diam est, varius id ullamcorper a, efficitur nec libero”
Last name First name
Position – Company
Success stories
Production quality project
Project to improve medical device manufacturing
EU Sensapnea project on sleep apnea
Project to improve patient data analysis
Project Eye Tracking
Project on eye tracking technology for human-machine interaction
Covid XRay
Open access scripts on model explainability for the healthcare field