Related Experiment Video
Updated: Aug 15, 2025
15:48
Tracking the Mammary Architectural Features and Detecting Breast Cancer with Magnetic Resonance Diffusion Tensor Imaging
Published on: December 15, 2014
22.6K
ETECADx: Ensemble Self-Attention Transformer Encoder for Breast Cancer Diagnosis Using Full-Field Digital X-ray
Aymen M Al-Hejri1,2, Riyadh M Al-Tam1,2, Muneer Fazea3,4
1School of Computational Sciences, Swami Ramanand Teerth Marathwada University, Nanded 431606, Maharashtra, India.
Diagnostics (Basel, Switzerland)
|January 8, 2023
Summary
A new AI framework, ETECADx, enhances early breast cancer detection by fusing convolutional neural networks and vision transformers. This advanced computer-aided diagnosis system achieves high accuracy, aiding radiologists in identifying malignancies.
Area of Science:
- Artificial Intelligence in Medical Imaging
- Oncology
- Computer-Aided Diagnosis
Background:
- Early breast cancer detection is critical for reducing mortality rates.
- Existing computer-aided diagnosis (CAD) systems can be improved with advanced AI techniques.
- Integrating deep learning and attention mechanisms offers potential for enhanced diagnostic accuracy.
Purpose of the Study:
- To propose a novel AI-based CAD framework, ETECADx, for early breast cancer detection.
- To fuse ensemble transfer learning from convolutional neural networks (CNNs) with the self-attention mechanism of a vision transformer encoder (ViT).
- To evaluate the framework's performance in both binary and multi-class classification of breast cancer.
Main Methods:
- Developed ETECADx by combining CNN ensemble learning for feature extraction and ViT for diagnosis.
- Utilized the INbreast dataset for training and validation, supplemented with private annotated breast cancer images.
- Evaluated performance using binary and multi-class classification approaches.
Main Results:
- Achieved high accuracies on the INbreast dataset: 98.58% (binary) and 97.87% (multi-class).
- ETECADx demonstrated significant improvements over individual backbone networks (6.6% binary, 4.6% multi-class).
- Validated on real breast images with accuracies of 97.16% (binary) and 89.40% (multi-class), with rapid prediction (0.048s/image).
Conclusions:
- The proposed ETECADx framework effectively improves breast cancer detection accuracy.
- The hybrid approach combining CNNs and ViT offers a robust solution for computer-aided diagnosis.
- ETECADx shows promise for practical applications, providing a valuable second opinion for radiologists.
