Skip to main content
Fig. 6 | Cancer Imaging

Fig. 6

From: Predicting microvascular invasion in hepatocellular carcinoma: a deep learning model validated across hospitals

Fig. 6

The developed ResNet-18 model for AP image and CF learned MVI relevant features that were similar to the conditions used by radiologists in estimating MVI clinically. Representative examples of attention heatmaps were generated by using the gradient-weighted class activation mapping (Grad-CAM) method for (A-B) true positive, (C-D) false positive, and (E-F) false negative in the validation and external sets. Heatmaps are standard jet colormaps and overlapped on the original input image. Red arrows indicate presence of internal arteries and black arrows indicate presence of hypodense halo. The actual MVI status based on the histopathological results, radiologist’s MVI estimation and MVI prediction of the ResNet-18 model are shown below each representative example. (G) An illustration of the clinical decision workflow applied by radiologists in estimating MVI

Back to article page