Objective: Helicobacter pylori (H. pylori) gastritis is strongly associated with gastric carcinogenesis. This study aimed to generate a prototype artificial intelligence (AI) diagnostic system for classifying H. pylori infection, using gastric X-ray images and deep learning(DL).
Subjects and Methods: First, 300 examinees (150 H. pylori positives) who underwent gastric X-ray examinations and serum H. pylori antibody tests were registered as a learning group to create an X-ray AI using DL. Subsequently, 100 examinees (50 H. pylori positives) from another hospital, with the same conditions as the learning group, were registered as a test group to evaluate the diagnostic accuracy of the AI. To evaluate its accuracy, an H. pylori serum antibody titer was set as the gold standard for infection diagnosis.
Results: The diagnostic ability of the AI was evaluated by a receiver operating characteristic curve. The diagnostic ability of H. pylori infection by AI was: sensitivity 76%, specificity 84%, diagnostic accuracy 80%, and area under the curve 0.82.
Conclusions: The gastric X-ray AI using DL demonstrated a relatively good ability to diagnose H. pylori infection. Subsequent studies can consider increasing the number of AI learning images along with research on the management of patients after the eradication of H. pylori.
We examined the frequency and characteristics of cases presenting with gastric cobblestone-like lesions (GCSLs) during endoscopic examinations at our facility. Of the 4,750 esophagogastroduodenoscopies, 15 (0.32%) demonstrated GCSLs. A history of proton pump inhibitor(PPI) usage and smoking were investigated as possible risk factors associated with GCSLs.
Among the 15 patients with GCSLs, 7 (47%) had a history of PPI usage and 13 patients (87%) reported a history of smoking. The male-to-female ratio was 13:2, and the mean age was 52.6 years. The frequency of GCSLs was significantly higher among smokers than nonsmokers. Serological Helicobacter pylori (H. pylori) antibody test results were negative for 11 patients. The remaining 4 patients did not undergo H. pylori testing but appeared to be H. pylori-negative based on their endoscopic findings. Furthermore, the endoscopic findings did not differ between patients with and without a past medical history of PPI usage and/or smoking.
Endoscopically, GCSLs are frequently observed in the lesser curvature of the body. Each nodule was small to medium in size. These features are important for detecting GCSLs. Moreover, the use of a small air volume during endoscopy is also important for identifying GCSLs.
Gastric radiography was performed in three cases, but the findings were minimal. Further explorations of gastric radiography in this context should be considered in the future.
Traditionally, GCSLs have been associated mainly with PPI usage. However, the findings from this study suggest that smoking may also be a significant factor.
The patient was a 72-year-old man. A blood test performed during his annual physical examination revealed a high CA19-9 level of 62.2 U/mL. As a CT scan revealed a tumor in the pancreatic head, he was referred to the hospital for further examination and treatment. A contrast-enhanced CT scan of his abdomen revealed a hypovascular tumor in the pancreatic head, which was suspected to be pancreatic cancer. An endoscopic ultrasonographic puncture aspiration biopsy was performed on the same area and the tumor was pathologically diagnosed as adenocarcinoma. Based on the procedure above, a diagnosis of pancreatic head cancer was made. The patient underwent a pancreaticoduodenectomy after preoperative chemotherapy. A histopathological examination of the surgical specimen revealed a honeycomb-like proliferation of poorly atypical cells with small round nuclei and pale sporangia. A poorly differentiated adenocarcinoma component was identified through hematoxylin eosin staining; both synaptophysin and chromogranin staining were shown to be positive through immunostaining. Part of the full-blown follicular structure was diagnosed as a neuroendocrine neoplasm. Since adenocarcinoma and neuroendocrine neoplasms were present in more than 30% of each other, the final diagnosis was primary mixed neuroendocrine-non-neuroendocrine neoplasm of the pancreatic head.