Related Experiment Video
Updated: Feb 19, 2026
07:15
Machine Learning Algorithms for Early Detection of Bone Metastases in an Experimental Rat Model
Published on: August 16, 2020
7.6K
Different BI-RADS breast cancer diagnosis using MobileNetV1 and vision transformer based on explainable artificial
Israa Abdelsabour1, Ahmed Elgarayhi1, Mohammed Sallah1
1Physics Department, Faculty of Science, Mansoura University, Mansoura, 35516, Egypt.
Scientific Reports
|February 17, 2026
Summary
A new hybrid deep learning framework combining MobileNetV1 and Vision Transformer (ViT) achieves over 99% accuracy for breast cancer (BC) classification from mammograms. This interpretable AI tool enhances early BC detection and supports clinical decisions.
Area of Science:
- Artificial Intelligence
- Medical Imaging
- Machine Learning
Background:
- Breast cancer (BC) remains a significant global health concern, necessitating advanced diagnostic tools.
- Current computer-aided diagnosis systems (CADs) require improvements in precision, effectiveness, and interpretability.
- Accurate BC classification from mammograms is crucial for timely intervention and improved patient outcomes.
Purpose of the Study:
- To introduce a novel hybrid deep learning (DL) framework for multi-class BI-RADS breast cancer classification using mammographic images.
- To enhance the accuracy and interpretability of BC diagnosis through a dual-stream network architecture.
- To develop an effective and explainable AI solution for clinical decision support in breast cancer detection.
Main Methods:
- A hybrid DL framework fusing MobileNetV1 (for local features) and Vision Transformer (ViT) (for global context) was developed.
- The framework employed feature-level fusion followed by a bagging-based logistic regression (LR) classifier for enhanced robustness.
- Evaluation utilized the King Abdulaziz University BC Mammogram Dataset (KAUBC) with 5-fold cross-validation and compared against state-of-the-art models.
- Explainable AI (XAI) techniques (Grad-CAM, Grad-CAM++) were used for visual interpretability.
Main Results:
- The proposed MobileNetV1-ViT-Bagging framework achieved high and stable performance across all BI-RADS categories.
- Accuracy (ACC), sensitivity (SEN), and specificity (SPE) exceeded 99% in classifying breast cancer.
- XAI techniques successfully highlighted diagnostically relevant regions, providing visual explanations for the model's predictions.
Conclusions:
- The MobileNetV1-ViT-Bagging framework offers an effective, computationally structured, and explainable solution for multi-class BI-RADS BC diagnosis.
- The hybrid approach demonstrates strong potential for clinical decision-support applications in mammography.
- The study highlights the efficacy of fusing CNNs and transformers for improved medical image analysis and interpretability.
