Fig. 4: Visualization of heatmaps generated by models. | Nature Communications

Fig. 4: Visualization of heatmaps generated by models.

From: An annotation-free whole-slide training approach to pathological classification of lung cancer types using deep learning

Fig. 4

Both heatmaps are upscaled to fit the original image size using a bicubic interpolation. The colors of the overlaid heatmaps represent the predicted probabilities of being a tumor, as defined in the color bar. a Whole-slide view on a slide containing adenocarcinoma lesions. b Heatmap of MIL on a whole-slide view. c Heatmap of whole-image method. d Zoom-in view with human annotations (blue contours). e Zoom-in view of MIL heatmap. f Zoom-in view of heatmap generated by whole-slide training method.

Back to article page