Deep Learning diagnosis of <i>Helicobacter pylori</i> infection using gastric X-ray images

DOI Web Site

Bibliographic Information

Other Title
  • Deep Learningと胃Ｘ線画像を用いた<i>Helicobacter pylori</i> 感染診断
  • Deep Learningと胃X線画像を用いたHelicobacter pylori感染診断
  • Deep Learning ト イ X センガゾウ オ モチイタ Helicobacter pylori カンセン シンダン

Search this article

Description

<p>AIMS: Diagnosis of Helicobacter pylori (H.pylori) infectious gastritis, a risk factor for gastric cancer, is established by serum antibodies as well as by visual evaluation of gastric X-ray double-contrast images. The use of X-ray images allows discrimination of active gastritis from non-gastritis by observing mucosal features and fold enlargement. Recent years have seen great advances in machine learning, including technologies that apply Deep Learning to medical images. Deep Learning is a method of machine learning that imitates neural cell physiology and is excellent for image and speech recognition applications. We aimed to determine whether active gastritis could be distinguished from non-gastritis from X-ray images by using Deep Learning.</p><p>METHODS: Overall, 100 patients underwent gastric X-ray examinations and serum H.pylori antibody tests on the same day. Serum antibody values were applied as the “gold standard” for diagnosing H.pylori infection, with antibody values <3 U/mL classified as H.pylori negative (n = 50) and those >10 U/mL classified as H.pylori positive (n = 50). Images obtained from both groups were sorted into training group of 70 persons (H.pylori positive; 35 persons) and image testing group of 30 persons (H.pylori negative; 15 persons). Each group was used for Deep Learning and evaluation of diagnostic ability. ALEXNET was adopted as the Deep Learning model. A receiver operating characteristic curve (ROC curve) was used to evaluate diagnostic accuracy.</p><p>RESULTS: The ROC curve indicated a sensitivity of 86.7% and a specificity of 91.7%. The area under the ROC curve was 0.921.</p><p>CONCLUSION: It is possible to diagnose H.pylori infection from gastric X-ray images using Deep Learning. Additionally, this method is useful for objective diagnosis of gastric cancer risk using X-ray images and auxiliary reading of the local diagnosis.</p>

Journal

Keywords

Details 詳細情報について

Report a problem

Back to top