TY  - JOUR
AU  - Hamza, Muhammad
AU  - Ali, Ibrar
AU  - Ali, Sikandar
AU  - Khan, Shayan
PY  - 2026
DA  - 2026/05/31
TI  - Multi-Scale CSPResNet50 with Feature Pyramid Aggregation and SE Attention for Breast Cancer Histopathological Subtype Classification and Malignancy Detection
JO  - ICCK Journal of Image Analysis and Processing
T2  - ICCK Journal of Image Analysis and Processing
JF  - ICCK Journal of Image Analysis and Processing
VL  - 2
IS  - 3
SP  - 122
EP  - 140
DO  - 10.62762/JIAP.2026.746947
UR  - https://www.icck.org/article/abs/JIAP.2026.746947
KW  - breast cancer
KW  - histopathology
KW  - CSPResNet50
KW  - GradCAM
KW  - deep learning
KW  - classification
AB  - Breast cancer is a leading cause of cancer-related mortality worldwide, making accurate histopathological subtype discrimination critical for timely clinical intervention. Existing deep learning approaches often evaluate limited settings (binary or multi-class, single magnification), restricting comparative utility and clinical interpretability. This study proposes a unified Cross Stage Partial Network (CSPNet)-based framework for comprehensive classification on the BreaKHis dataset. A CSPResNet50 backbone pre-trained on ImageNet was extended with a multi-scale Feature Pyramid-style aggregation head, Squeeze-and-Excitation channel attention, dual Global Average and Max Pooling per scale (1536-dimensional descriptor), and a dense classification neck, trained via two-phase progressive fine-tuning. Four tasks were evaluated under a stratified 70/15/15 split: multi-class (8 subtypes) and binary classification at whole-dataset and per-magnification levels (40×, 100×, 200×, 400×), with GradCAM providing interpretable visualisations. For whole-dataset binary classification, the model achieves 95.53% accuracy, 94.80% F1-score, 96.93% sensitivity, and 98.53% AUC. In multi-class classification, it attains 78.10% accuracy, 81.40% balanced accuracy, 76.76% macro F1-score, and 97.72% AUC. Per-magnification analysis shows best performance at 40×, with binary and multi-class accuracy reaching 95.00% and 77.00%, respectively, while AUC remains above 92% across all magnifications. A key limitation is image-level rather than patient-level splitting, which may introduce optimistic bias. Overall, the proposed framework provides a robust, interpretable, and computationally efficient solution for breast cancer histopathological image classification.
SN  - 3068-6679
PB  - Institute of Central Computation and Knowledge
LA  - English
ER  - 

@article{Hamza2026MultiScale,
  author = {Muhammad Hamza and Ibrar Ali and Sikandar Ali and Shayan Khan},
  title = {Multi-Scale CSPResNet50 with Feature Pyramid Aggregation and SE Attention for Breast Cancer Histopathological Subtype Classification and Malignancy Detection},
  journal = {ICCK Journal of Image Analysis and Processing},
  year = {2026},
  volume = {2},
  number = {3},
  pages = {122-140},
  doi = {10.62762/JIAP.2026.746947},
  url = {https://www.icck.org/article/abs/JIAP.2026.746947},
  abstract = {Breast cancer is a leading cause of cancer-related mortality worldwide, making accurate histopathological subtype discrimination critical for timely clinical intervention. Existing deep learning approaches often evaluate limited settings (binary or multi-class, single magnification), restricting comparative utility and clinical interpretability. This study proposes a unified Cross Stage Partial Network (CSPNet)-based framework for comprehensive classification on the BreaKHis dataset. A CSPResNet50 backbone pre-trained on ImageNet was extended with a multi-scale Feature Pyramid-style aggregation head, Squeeze-and-Excitation channel attention, dual Global Average and Max Pooling per scale (1536-dimensional descriptor), and a dense classification neck, trained via two-phase progressive fine-tuning. Four tasks were evaluated under a stratified 70/15/15 split: multi-class (8 subtypes) and binary classification at whole-dataset and per-magnification levels (40×, 100×, 200×, 400×), with GradCAM providing interpretable visualisations. For whole-dataset binary classification, the model achieves 95.53\% accuracy, 94.80\% F1-score, 96.93\% sensitivity, and 98.53\% AUC. In multi-class classification, it attains 78.10\% accuracy, 81.40\% balanced accuracy, 76.76\% macro F1-score, and 97.72\% AUC. Per-magnification analysis shows best performance at 40×, with binary and multi-class accuracy reaching 95.00\% and 77.00\%, respectively, while AUC remains above 92\% across all magnifications. A key limitation is image-level rather than patient-level splitting, which may introduce optimistic bias. Overall, the proposed framework provides a robust, interpretable, and computationally efficient solution for breast cancer histopathological image classification.},
  keywords = {breast cancer, histopathology, CSPResNet50, GradCAM, deep learning, classification},
  issn = {3068-6679},
  publisher = {Institute of Central Computation and Knowledge}
}