Extended Data Fig. 2: Explainability Analysis of DeepRETStroke System in Detecting Silent Brain Infarction.
From: A deep learning system for detecting silent brain infarction and predicting stroke risk
Left Panel: Original color fundus images of participants. Middle Panel: SHAP (Shapley Additive Explanations) summary plot of fundus images. The pixel color from green to blue indicates the increasing expectation of gradients for SBI detection. Right Panel: Occlusion-based attribution plot of fundus images. Red pixels reveal the areas with positive attribution highlighting possible pathological changes. For the detection of SBI (silent brain infarct), some anatomical structures associated with specific diseases, for example retinal vasculature, are highlighted. MRI Image: MRI lesion of SBI was highlighted.
