Figure 3 | Scientific Reports

Figure 3

From: Deep learning systems detect dysplasia with human-like accuracy using histopathology and probe-based confocal laser endomicroscopy

Figure 3

Examples of pCLE images from each class (top row) and gradient-weighted Class Activation Maps (Grad-CAMs) that highlight regions of pCLE frames relevant to classification by the convolutional neural network models (bottom row). See 20-micron reference bar for resolution (bottom right). Red/orange coloration indicates high activation while blue/green coloration indicates low activation. Left column: an example of squamous tissue with the Grad-CAM clearly focusing on the intrapapillary loop in the center as the region of interest. Middle column: an example of Barrett’s esophagus with the goblet cells and columnar epithelium being highlighted as the regions of interest. Right column: an example of dysplasia with the Grad-CAM focusing on the poorly organized, saw-toothed epithelium without clear goblet cells that is indicative of dysplasia. Created using matplotlib 3.3.2 (https://matplotlib.org/).

Back to article page