Related Experiment Video
Updated: Aug 24, 2025
03:31
Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications
Published on: December 15, 2023
606
Improving Performance of Breast Lesion Classification Using a ResNet50 Model Optimized with a Novel Attention
Warid Islam1, Meredith Jones2, Rowzat Faiz1
1School of Electrical & Computer Engineering, University of Oklahoma, Norman, OK 73019, USA.
Tomography (Ann Arbor, Mich.)
|October 26, 2022
Summary
A new attention mechanism improved deep transfer learning models for breast lesion classification on mammograms, significantly boosting accuracy in distinguishing malignant from benign cases.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Machine Learning
Background:
- Accurate classification of malignant vs. benign breast lesions on mammograms is vital for effective breast cancer screening.
- Current methods face challenges in reducing false positives and improving screening efficacy.
Purpose of the Study:
- To optimize a deep transfer learning model for breast lesion classification.
- To enhance classification accuracy by implementing a novel attention mechanism.
Main Methods:
- ResNet50 was selected as the base model for a new deep transfer learning approach.
- A convolutional block attention module (CBAM) was integrated into ResNet50.
- A dataset of 4280 mammograms with benign and malignant lesions was used for training and testing.
Main Results:
- The CBAM-based ResNet50 model achieved an Area Under the ROC Curve (AUC) of 0.866 ± 0.015.
- This performance was significantly higher than the standard ResNet50 model (AUC = 0.772 ± 0.008).
Conclusions:
- Optimizing deep transfer learning models with attention mechanisms can significantly improve performance in medical imaging tasks.
- This approach holds promise for enhancing the accuracy of breast cancer screening.
