Fig. 5 | Scientific Reports

Fig. 5

From: Noncontrast CT-based deep learning for predicting intracerebral hemorrhage expansion incorporating growth of intraventricular hemorrhage

Fig. 5

Visualization of attention maps for the 2D-ResNet-101 model. The gradient-weighted class activation mapping (Grad-CAM) of the 2D-ResNet-101 deep learning model in intracerebral hemorrhage patients with revised hematoma expansion (rHE) (A) and non-rHE (B). To visually verify the decision-making process of the 2D-ResNet-101, Grad-CAM was applied to generate attention maps. These maps were then overlaid onto the input images, producing final overlay attention maps that highlight the hematoma regions most impacting the model predictions. The red regions indicate areas most influential to the model’s classification, primarily located at the hematoma and its periphery regions.

Back to article page