Related Experiment Video
Updated: Oct 7, 2025
06:03
Integrating Augmented Reality Tools in Breast Cancer Related Lymphedema Prognostication and Diagnosis
Published on: February 6, 2020
6.7K
A bimodal BI-RADS-guided GoogLeNet-based CAD system for solid breast masses discrimination using transfer learning
Zahra Assari1, Ali Mahloojifar1, Nasrin Ahmadinejad2
1Department of Biomedical Engineering, Faculty of Electrical and Computer Engineering, Tarbiat Modares University, Tehran, Iran.
Computers in Biology and Medicine
|January 7, 2022
Summary
This study introduces a novel bimodal computer-aided diagnosis (CAD) system that combines mammography and ultrasound images for solid breast mass classification. The system significantly improves diagnostic accuracy for breast cancer detection.
Area of Science:
- Medical Imaging
- Artificial Intelligence in Medicine
- Oncology
Background:
- Accurate diagnosis of solid breast masses is crucial for effective treatment planning.
- Integrating information from multiple imaging modalities like mammography and ultrasound presents a diagnostic challenge for radiologists.
- Existing computer-aided diagnosis (CAD) systems often analyze imaging data from a single modality, limiting diagnostic potential.
Purpose of the Study:
- To develop and evaluate a novel bimodal CAD system for solid breast mass classification using both mammographic and sonographic images.
- To enhance diagnostic performance by effectively combining complementary information from different imaging modalities.
- To leverage deep learning techniques, specifically GoogLeNet, for improved accuracy in breast cancer detection.
Main Methods:
- A bimodal GoogLeNet-based CAD system was developed, training distinct monomodal models first, followed by a bimodal model using high-level feature maps.
- Image content representations were optimized to exploit BI-RADS descriptors for comprehensive analysis.
- A two-step transfer learning strategy was employed using an ImageNet pre-trained GoogLeNet model and multiple datasets.
Main Results:
- The bimodal model achieved high performance metrics: 90.91% sensitivity, 89.87% specificity, 90.32% F1-score, 80.78% Matthews Correlation Coefficient, 95.82% AUC, and 90.38% accuracy.
- The proposed system demonstrated superior recognition results compared to monomodal approaches.
- The integration of mammographic and sonographic data significantly enhanced the classification of solid breast masses.
Conclusions:
- The novel bimodal CAD system effectively integrates information from mammography and ultrasound for solid breast mass classification.
- The system shows significant potential to improve breast cancer diagnostic performance and assist radiologists.
- This approach offers a promising direction for developing advanced AI tools in breast imaging analysis.
