Transformer-Based Multiple Instance Learning with Gated Feature Fusion for Gastric Cancer Detection from Histopathological Images
International Journal of Drug Delivery Technology · 30.05.2026 · 16 (38s)
DOI: 10.25258/ijddt.16.38s.4 ↗Kurz gefasst
Fachartikel in der Zeitschrift International Journal of Drug Delivery Technology, online erschienen am 30.05.2026. Verfasst von Harshal Diwakarrao Wankhade und 1 weiteren Autor:innen.
Abstract (Originalsprache)
Gastric cancer is the most common cause of cancer-related deaths globally, and early diagnosis through histological examination is essential to increase patient survival rates. However, manual interpretation of histopathological slides is time-consuming and subject to inter-observer variability, particularly when analyzing large numbers of wholeslide images. In recent years, deep learning techniques show significant potential for automated cancer detection, and many existing convolutional neural network models only focused on local spatial features and fail to capture contextual relationships among multiple tissue regions. This study proposes a Transformer-based Multiple Instance Learning model with Gated Feature Fusion for gastric cancer detection from histopathological images. The proposed model combined the EfficientNetB0-based convolutional feature extraction, transformer-based contextual modeling, and a gated fusion mechanism to selectively discriminative features and remove the redundant data. Experiments were conducted on a publicly available gastric histopathology dataset. The proposed model obtained the accuracy score of 0.990, precision of 0.9903, recall of 0.9895, F1-score of 0.9895, MCC of 0.9887, and Kappa of 0.9886 that performed well as compared to baseline CNN and hybrid models. The study also conducted the statistical significance test using McNemar’s test, bootstrap confidence intervals, and permutation analysis that indicate the robustness and reliability of the proposed model. The final results shows that the TransMIL-GFN model indicate the effective and reliable approach for automated gastric cancer classification and has strong potential to support computer-assisted histopathological diagnosis in clinical practice.
Zitieren
APA
Wankhade, H. D., & Zade, A. V. (2026). Transformer-Based Multiple Instance Learning with Gated Feature Fusion for Gastric Cancer Detection from Histopathological Images. International Journal of Drug Delivery Technology, 16(38s). https://doi.org/10.25258/ijddt.16.38s.4
Vancouver
Wankhade HD, Zade AV. Transformer-Based Multiple Instance Learning with Gated Feature Fusion for Gastric Cancer Detection from Histopathological Images. International Journal of Drug Delivery Technology. 2026;16(38s). doi:10.25258/ijddt.16.38s.4
BibTeX
@article{Wankhade2026,
author = {Wankhade, Harshal Diwakarrao and Zade, Amol V.},
title = {Transformer-Based Multiple Instance Learning with Gated Feature Fusion for Gastric Cancer Detection from Histopathological Images},
journal = {International Journal of Drug Delivery Technology},
year = {2026},
volume = {16},
number = {38s},
doi = {10.25258/ijddt.16.38s.4}
}
EUVIMED