An Efficient Hybrid ResNet-101 with CBAM and Bi-LSTM for Breast Cancer Recognition
International Journal on Advanced Science, Engineering and Information Technology · 25.10.2026 · 16 (5) · 1896-1907
DOI: 10.18517/ijaseit.16.5.14843 ↗Kurz gefasst
Fachartikel in der Zeitschrift International Journal on Advanced Science, Engineering and Information Technology, online erschienen am 25.10.2026. Verfasst von Kashif Khan und 8 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
Computer Aided Diagnosis systems often struggle to attain high classification accuracy due to small morphological differences between benign tumors. This challenge arises from the limited ability to extract and preserve discriminative features throughout the deep learning pipeline. To address this, we propose an enhanced ResNet101V2 architecture integrated with a Convolutional Block Attention Module (CBAM) and a Bidirectional Long Short-Term Memory (BiLSTM) module. ResNet101V2 extracts discriminative features, CBAM emphasizes salient regions, and BiLSTM captures contextual dependencies to enrich feature representations. The proposed model was evaluated using BreakHis dataset across individual magnification levels (40X, 100X, 200X, and 400X) and combined dataset settings using both image-wise split and patient-wise 5-fold cross-validation split. The model outperforms existing approaches, with precision ranging from 96.88% to 98.88%, recall from 88.76% to 98.94%, AUC from 98.38% to 99.93%, F1-score from 0.9462 to 0.9840, and accuracy from 94.85% to 98.40%. For the patient-wise split, the average accuracy ranges from 0.8415 to 0.9138, precision from 0.8513 to 0.9176, recall from 0.8933 to 0.9499, F1-score from 0.8781 to 0.9325 and AUC from 0.9468 to 0.9468. The findings indicate the model's strong performance in capturing and preserving discriminative features, thereby enhancing classification performance and improving patient outcomes. The process innovation lies in sequentially integrating deep residual feature extraction, attention-based salient region refinement, and bidirectional contextual modeling to progressively preserve and enhance discriminative tumor characteristics throughout the classification pipeline.
Zitieren
APA
Khan, K., Awang, S., Talab, M. A., Nafis, S. M., Khan, I., Onn, C. W., Kuuanishbaevna, B. A., Kasim, S., & Arshad, M. S. (2026). An Efficient Hybrid ResNet-101 with CBAM and Bi-LSTM for Breast Cancer Recognition. International Journal on Advanced Science, Engineering and Information Technology, 16(5), 1896-1907. https://doi.org/10.18517/ijaseit.16.5.14843
Vancouver
Khan K, Awang S, Talab MA, Nafis SM, Khan I, Onn CW, et al. An Efficient Hybrid ResNet-101 with CBAM and Bi-LSTM for Breast Cancer Recognition. International Journal on Advanced Science, Engineering and Information Technology. 2026;16(5):1896-1907. doi:10.18517/ijaseit.16.5.14843
BibTeX
@article{Khan2026,
author = {Khan, Kashif and Awang, Suryanti and Talab, Mohammad Ahmed and Nafis, Syafiqah Mohd and Khan, Iqra and Onn, Choo Wou and Kuuanishbaevna, Bazarbaeva Aygul and Kasim, Shahreen and Arshad, Mohamad Syafwan},
title = {An Efficient Hybrid ResNet-101 with CBAM and Bi-LSTM for Breast Cancer Recognition},
journal = {International Journal on Advanced Science, Engineering and Information Technology},
year = {2026},
volume = {16},
number = {5},
pages = {1896--1907},
doi = {10.18517/ijaseit.16.5.14843}
}
EUVIMED