Binary Secretary Bird Optimization-Based Deep Feature Selection for Explainable Lung Cancer Histopathology Classification Using ResNet50-SE and Test-Time Augmentation
Journal of Sustainable Research in Applied Sciences · 01.10.2026 · 3 (2) · 143-159
DOI: 10.36602/jsras.2026.3.2.36 ↗Kurz gefasst
Fachartikel in der Zeitschrift Journal of Sustainable Research in Applied Sciences, online erschienen am 01.10.2026. Verfasst von Amna Ali A Mohamed. Die Arbeit nennt 47 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC-SA 4.0).
Abstract (Originalsprache)
Accurate histopathological subtyping is critical for lung cancer treatment planning, yet manual slide analysis remains time-consuming and prone to significant inter-observer variability. This paper proposes an explainable hybrid deep learning framework for automated lung cancer histopathological image classification. The framework integrates a ResNet50 backbone with Squeeze-and-Excitation (SE) channel attention and selective fine-tuning, a Binary Secretary Bird Optimization Algorithm (SBOA) for deep feature selection, Test-Time Augmentation (TTA) for inference-time feature stabilization, a multi-classifier evaluation stage comparing K-Nearest Neighbor (KNN), Support Vector Machine (SVM), and Decision Tree classifiers, and Gradient-weighted Class Activation Mapping (Grad-CAM) for visual explainability. On the LC25000 lung histopathology benchmark, Binary SBOA reduced the 256-dimensional feature vector by 90.2% retaining only 25 features, while KNN achieved 99.78% accuracy and a mean of 99.63% ± 0.14% across five independent seeds. SVM and Decision Tree achieved single-run accuracies of 99.56% and 98.80%, with mean robustness of 96.35% ± 2.96% and 98.93% ± 0.09% across five seeds, respectively. A lightweight MobileNetV2-SE variant achieved 97.60% accuracy with 87% fewer parameters. Grad-CAM attention appears consistent with diagnostically relevant regions, pending pathologist validation. To the best of the authors' knowledge, this is the first application of SBOA to deep feature selection in histopathological image classification, demonstrating that bio-inspired optimization combined with attention-modulated deep features yields an accurate, computationally efficient, and clinically interpretable diagnostic framework
Zitieren
APA
Mohamed, A. A. A. (2026). Binary Secretary Bird Optimization-Based Deep Feature Selection for Explainable Lung Cancer Histopathology Classification Using ResNet50-SE and Test-Time Augmentation. Journal of Sustainable Research in Applied Sciences, 3(2), 143-159. https://doi.org/10.36602/jsras.2026.3.2.36
Vancouver
Mohamed AAA. Binary Secretary Bird Optimization-Based Deep Feature Selection for Explainable Lung Cancer Histopathology Classification Using ResNet50-SE and Test-Time Augmentation. Journal of Sustainable Research in Applied Sciences. 2026;3(2):143-159. doi:10.36602/jsras.2026.3.2.36
BibTeX
@article{Mohamed2026,
author = {Mohamed, Amna Ali A},
title = {Binary Secretary Bird Optimization-Based Deep Feature Selection for Explainable Lung Cancer Histopathology Classification Using ResNet50-SE and Test-Time Augmentation},
journal = {Journal of Sustainable Research in Applied Sciences},
year = {2026},
volume = {3},
number = {2},
pages = {143--159},
doi = {10.36602/jsras.2026.3.2.36}
}
EUVIMED