Skip to main content
. 2019 Jul 1;26(11):1286–1296. doi: 10.1093/jamia/ocz089

Figure 6.

Figure 6.

Confusion matrices of various models and depths. Each confusion matrix is color-coded as a heatmap for visualization purpose. Models 3 and 4 (Convolutional AutoEncoder based Semi-supervised Network) consistently achieves better performance compared with model 2 at all network depths. Duod: duodenum; GasAnt: gastric antrum; GasBody: gastric body; GasCard: gastric cardia; HighG: high-grade dysplasia; IneMet: intestinal metaplasia; IntraCar: intraepithelial carcinoma; LowG: low-grade dysplasia; Sq: squamous.