EUVIMEDEUVIMED entdecken
Open Access · CC BY-NC 3.0

Development of a Pipeline Machine Learning Model to Identify and Count Helicobacter pylori in Gastric Biopsies

Fang Wang, Shuangyan Lin, Yanbo Lv, Luoning Zheng, Yixin Chen, Fang Peng, Zhifeng Shuai

International Surgery · 2026-04 · 110 (1) · 25-32

DOI: 10.9738/intsurg-d-25-00045.1 ↗

Kurz gefasst

Fachartikel in der Zeitschrift International Surgery, 2026-04. Verfasst von Fang Wang und 6 weiteren Autor:innen. Die Arbeit nennt 46 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC 3.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Objective

Helicobacter pylori (Hp) infection affects half of the global population, with rising antibiotic resistance causing >20% treatment failure. Clinically, gastric biopsy is routinely performed in patients with gastrointestinal symptoms. On hematoxylin and eosin (H&E)-stained histological images, lesion characteristics can be assessed to guide therapy, whereas Hp can also be identified.

Methods

We promote the development of precise augmentation and counting of Hp (PAC-Hp), a deep learning framework combining density-based population counting algorithms with synthetic data augmentation using deep convolutional generative adversarial networks (DCGANs). The system automates Hp identification and quantification directly from histological images. We trained and validated PAC-Hp using whole-slide images (WSIs) from 114 patients, using immunohistochemistry (IHC) and multiplex immunofluorescence (mIF) as the gold standard for Hp detection.

Results

PAC-Hp achieved superior performance (88.6% sensitivity, 89.4% specificity) compared to manual microscopic diagnosis (82.4% sensitivity, 88.1% specificity). The framework demonstrated robust accuracy in quantifying Hp density while mitigating challenges like staining variability and overlapping bacteria.

Conclusions

Our artificial intelligence (AI)-driven approach provides a cost-effective, scalable solution for Hp detection, reducing reliance on labor-intensive manual methods. PAC-Hp’s high accuracy could aid in guiding therapy, particularly in resource-limited settings, and warrants further validation in multicenter studies.

Zitieren

APA

Wang, F., Lin, S., Lv, Y., Zheng, L., Chen, Y., Peng, F., & Shuai, Z. (2026). Development of a Pipeline Machine Learning Model to Identify and Count Helicobacter pylori in Gastric Biopsies. International Surgery, 110(1), 25-32. https://doi.org/10.9738/intsurg-d-25-00045.1

Vancouver

Wang F, Lin S, Lv Y, Zheng L, Chen Y, Peng F, et al. Development of a Pipeline Machine Learning Model to Identify and Count Helicobacter pylori in Gastric Biopsies. International Surgery. 2026;110(1):25-32. doi:10.9738/intsurg-d-25-00045.1

BibTeX
@article{Wang2026,
  author = {Wang, Fang and Lin, Shuangyan and Lv, Yanbo and Zheng, Luoning and Chen, Yixin and Peng, Fang and Shuai, Zhifeng},
  title = {Development of a Pipeline Machine Learning Model to Identify and Count Helicobacter pylori in Gastric Biopsies},
  journal = {International Surgery},
  year = {2026},
  volume = {110},
  number = {1},
  pages = {25--32},
  doi = {10.9738/intsurg-d-25-00045.1}
}
Kostenlos starten →