Related Experiment Video
Updated: Mar 15, 2026
10:37
A Multimodal Imaging Framework to Advance Phenotyping of Living Label-free Breast Cancer Cells
Published on: August 22, 2025
1.4K
Ensemble Deep Learning-Based High-Precision Framework for Breast Cancer Detection from Histopathological Images
Faizan Ahmad1, Arfan Jaffar1, Ghazanfar Latif2
1Department of Computer Science, Superior University, Lahore 547700, Pakistan.
Diagnostics (Basel, Switzerland)
|March 14, 2026
Summary
This study introduces a hybrid deep learning framework integrating Convolutional Neural Networks (CNNs) and Vision Transformers (ViTs) for enhanced breast cancer diagnosis from histopathological images. The proposed model achieves high accuracy and robustness, offering an interpretable solution for automated detection.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Computational Pathology
Background:
- Histopathological image analysis is standard for breast cancer diagnosis.
- Current deep learning and ViT architectures face challenges in capturing complex patterns, leading to overfitting and optimization issues.
Purpose of the Study:
- To propose a novel four-phase hybrid framework to enhance feature fusion for improved breast cancer diagnosis.
- To increase the strength, robustness, and generalization ability of diagnostic models.
Main Methods:
- A patient-wise data split (70-15-15) with augmentation and cross-validation was used.
- Independent training of CNNs (VGG16, ResNet50, DenseNet121) and ViTs (DeiT, CaiT, T2T-ViT, Swin Transformer) established baseline performance.
- A hybrid framework combined the best CNN and ViT models using self-attention for cross-modal feature fusion, Global Average Pooling, and feature scaling.
Main Results:
- The hybrid framework achieved 98.7% accuracy and 98.7% F1-score on the BreakHis dataset.
- The model demonstrated 95.8% accuracy on the external BACH dataset.
- XGBoost was the top-performing machine learning classifier, with results supported by Grad-CAM and Grad-CAM++ interpretability.
Conclusions:
- Integrating CNNs and ViTs via self-attention provides a robust and interpretable solution for automated breast cancer diagnosis.
- The proposed framework effectively addresses limitations of existing deep learning architectures.