EUVIMEDEUVIMED entdecken
Crossref

A lightweight shufflenet-V2 model for multi-class gastric histopathology classification with explainable AI

H. D. Wankhade, A. V. Zade

Nigerian Journal of Technology · 08.06.2026 · 45 · 87-95

DOI: 10.4314/njt.v45i.7s ↗

Kurz gefasst

Fachartikel in der Zeitschrift Nigerian Journal of Technology, online erschienen am 08.06.2026. Verfasst von H. D. Wankhade und 1 weiteren Autor:innen.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Gastric cancer is among the predominant causes of cancer-related death in the world that must be accurately and dependably examined by the histopathology. Proper interpretation of tissues is critical in early diagnosis in proper staging, and effective therapeutic decisions. Nevertheless, whole slide analysis using a manual method is tedious and subject to inter-rater error, whereas currently used deep learning systems tend to be computationally expensive and lack interpretability. The proposed paper uses a lightweight but a high-performing framework utilizing ShuffleNet-V2 to classify the gastric histopathology images into multi-classes. A subsample of 8000 histopathological image patches was taken out of a publicly accessible Kaggle dataset that comprises of 1000 samples of eight tissue classes. The proposed ShuffleNet-V2 with SE attention model was trained on dataset and compared with a number of pretrained models CNN, ResNet50, DenseNet201, EfficientNet-B3, MobileNet-V3 Large and Inception-V3 and identify best performing model. The proposed models were assessed in the different evaluation parameters. Also analyze the Mahalanobis distance, explainable AI and complemented by Mahalanobis distance. The model of ShuffleNet-V2 achieved an accuracy of 98.98. The ideal ROC and PR curves of the model are 99.90 and 99.90 respectively and high class-wise F1-score constantly, in comparison to all the other models. The explainable AI visualize the attribution maps that show the model to pay attention to the histologically significant structures, and favor meaningful clinical interpretability. The ultimate conclusion is that, ShuffleNet-V2 model is an accurate, efficient and explainable model to classify gastric tissue, and has high potential to be incorporated in the digital pathology process.

Zitieren

APA

Wankhade, H. D., & Zade, A. V. (2026). A lightweight shufflenet-V2 model for multi-class gastric histopathology classification with explainable AI. Nigerian Journal of Technology, 45, 87-95. https://doi.org/10.4314/njt.v45i.7s

Vancouver

Wankhade HD, Zade AV. A lightweight shufflenet-V2 model for multi-class gastric histopathology classification with explainable AI. Nigerian Journal of Technology. 2026;45:87-95. doi:10.4314/njt.v45i.7s

BibTeX
@article{Wankhade2026,
  author = {Wankhade, H. D. and Zade, A. V.},
  title = {A lightweight shufflenet-V2 model for multi-class gastric histopathology classification with explainable AI},
  journal = {Nigerian Journal of Technology},
  year = {2026},
  volume = {45},
  pages = {87--95},
  doi = {10.4314/njt.v45i.7s}
}
Kostenlos starten →