Skip to main content
Fig. 2 | BMC Gastroenterology

Fig. 2

From: Deep learning for sensitive detection of Helicobacter Pylori in gastric biopsies

Fig. 2

Validation strategy of the decision support algorithm. a Visualization of H. pylori detection in both Giemsa and H&E stained images. The green border highlights detected H. pylori bodies that the network correctly classified as H. pylori. b Illustration of the approach of validating the AI algorithm. About 347 Giemsa-stained slides (blue dots on illustrated slide) and 364 H&E slides (red dots on illustrated slide) were used, following an extraction of H. pylori hot spots. Then, the extracted hot spots were classified and ranked by H. pylori presence. The bean like structure of H. pylori is shown to visualize the scale. c The extracted and classified H. pylori hot spots were then annotated for H. pylori presence, as shown by a red box around four of the six tiles. The numbers correspond to an exemplified rank for H. pylori detection. For simplification, only Giemsa pictures are shown, while H&E stains were used for validation as well

Back to article page