Scientific Reports
Model assesses several health conditions from retinal images
In tests using four retinal image datasets, the model achieved 94.7% accuracy and outperformed models built for individual tasks.In tests using four retinal image datasets, the model achieved 94.7% accuracy and outperformed models built for individual tasks.
Computational studyEye
Who was studiedRetinal fundus images from multiple datasets, against single-task models
What changedThe model predicted health conditions with 94.7% accuracy, performing better than single-task models
The catchCardiovascular risk was an image-based surrogate rather than a validated clinical diagnosis
