Development of a scoring model for the Sharp/van der Heijde score using convolutional neural networks and its clinical application
Authors' conclusion
A high-performance scoring model for SHS that is applicable to clinical research was constructed.
Study context
- Population
- Patients with RA (in-house RA image database)
- Outcome
- Prediction of SHS (orientation, detection and damage prediction) by CNN model
Methodological detailsconvolutional neural networks, graph convolutional network, in-house RA image database
- Convolutional neural networks
- Graph convolutional network
- In-house RA image database
Main finding
As reported in the abstract
CNN model performance for SHS
- Image orientation correction accuracy
- 100%
- Predicted joint coordinates within 10 pixels of correct coordinates
- all
- κ between model and expert 1, erosion
- 0.879
- κ between model and expert 1, joint space narrowing
- 0.865
Observational data: associations, not effects.
Further findings
Graph convolutional network (GCN) model prediction of radiographic progression (ΔSHS >3/year)
- AUC, with biologics at baseline
- 0.753
- AUC, without biologics at baseline
- 0.734
- Superior to the other models
- RF titre the most important feature in biologics users
Clinical features associated with convolutional neural networks (CNN)-predicted ΔSHS per year in biologics users
- High-titre RF an independent risk factor
Original publication
Development of a scoring model for the Sharp/van der Heijde score using convolutional neural networks and its clinical application. Rheumatology (Oxford). Published 11 October 2022. doi:10.
Composed by AI from the published abstract and checked against it. How Briefs are made →