Application of Convolutional Neural Networks in the Diagnosis of Helicobacter pylori Infection Based on Endoscopic Images

Satoki Shichijo,Shuhei Nomura,Kazuharu Aoyama,Yoshitaka Nishikawa,Motoi Miura,Takahide Shinagawa,Hirotoshi Takiyama,Tetsuya Tanimoto,Soichiro Ishihara,Keigo Matsuo,Tomohiro Tada

doi:10.1016/j.ebiom.2017.10.014

Abstract

Background and aimsThe role of artificial intelligence in the diagnosis of Helicobacter pylori gastritis based on endoscopic images has not been evaluated. We constructed a convolutional neural network (CNN), and evaluated its ability to diagnose H. pylori infection. MethodsA 22-layer, deep CNN was pre-trained and fine-tuned on a dataset of 32,208 images either positive or negative for H. pylori (first CNN). Another CNN was trained using images classified according to 8 anatomical locations (secondary CNN). A separate test data set (11,481 images from 397 patients) was evaluated by the CNN, and 23 endoscopists, independently. ResultsThe sensitivity, specificity, accuracy, and diagnostic time were 81.9%, 83.4%, 83.1%, and 198s, respectively, for the first CNN, and 88.9%, 87.4%, 87.7%, and 194s, respectively, for the secondary CNN. These values for the 23 endoscopists were 79.0%, 83.2%, 82.4%, and 230±65min (85.2%, 89.3%, 88.6%, and 253±92min by 6 board-certified endoscopists), respectively. The secondary CNN had a significantly higher accuracy than endoscopists (by 5.3%; 95% CI, 0.3–10.2). ConclusionH. pylori gastritis could be diagnosed based on endoscopic images using CNN with higher accuracy and in a considerably shorter time compared to manual diagnosis by endoscopists.

Highlights

Gastric cancer is one of the most common malignancies, with one million cases estimated around the world in 2012 (O'Connor et al, 2017)
The convolutional neural network (CNN) constructed in this study provided an output of the probability of H. pylori infection per image
At a cut off value of 0.43, the value for which the point on the Receiver operating curves (ROC) curve corresponds to 100% sensitivity and specificity, the sensitivity, specificity, and accuracy of the CNN were 81.9% (95% confidence interval [CI], 71.1–90.0), 83.4%, and 83.1%, respectively

Summary

Introduction

Gastric cancer is one of the most common malignancies, with one million cases estimated around the world in 2012 (O'Connor et al, 2017). Patients with a positive test result are considered for H. pylori eradication therapy for the prevention of gastric cancer and other diseases, which are covered by national health insurance in Japan. The role of artificial intelligence in the diagnosis of Helicobacter pylori gastritis based on endoscopic images has not been evaluated. Methods: A 22-layer, deep CNN was pre-trained and fine-tuned on a dataset of 32,208 images either positive or negative for H. pylori (first CNN). Another CNN was trained using images classified according to 8 anatomical locations (secondary CNN). Conclusion: H. pylori gastritis could be diagnosed based on endoscopic images using CNN with higher accuracy and in a considerably shorter time compared to manual diagnosis by endoscopists

Methods

Results

Conclusion