Fig. 8.

a Transfer learning layout. A GoogleNet Inception v3 CNN, pre-trained on the ImageNet data is fine-tuned with CARS images comprising four classes: normal, small-cell carcinoma, squamous carcinoma, and adenocarcinoma. b–e Representative CARS images and f–i corresponding H&E-stained images of human lung tissues: b, f normal lung, c, g adenocarcinoma, d, h squamous cell carcinoma, and e, i small-cell carcinoma. Scale bars: 50 μm. Deep CNN model performance on the j normalized confusion matrix. Each row represents the instances in a ground-truth class and the value in each column represents what percentage of the images is predicted to a certain class. k ROC curves for three conditions: separating cancerous from normal lung images (light blue); separating small-cell carcinoma from nonsmall cell carcinoma lung images (dark blue); separating adenocarcinoma and squamous carcinoma lung images (orange). AUC scores are given in the legend. This figure is adapted with permission from Weng et al. 2017, SPIE