Related Experiment Video
Updated: Jul 8, 2025
07:15
Machine Learning Algorithms for Early Detection of Bone Metastases in an Experimental Rat Model
Published on: August 16, 2020
6.8K
Comparative analysis of deep learning methods for lesion detection on full screening mammography
Summary
Deep learning models like CenterNet and YOLOv5 show high accuracy in detecting breast cancer lesions from mammograms. These advanced techniques offer improved diagnostic performance for breast cancer screening.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Oncology
Background:
- Breast cancer is the most common cancer in women, with mammography as the primary diagnostic tool.
- Detecting lesions in mammograms is challenging due to poor contrast and varied lesion characteristics.
- Deep learning (DL) shows promise for improving automated medical image analysis.
Purpose of the Study:
- To benchmark state-of-the-art deep learning methods for breast lesion detection in mammography.
- To evaluate the performance of five leading DL architectures on a public dataset.
- To compare detection accuracy using metrics including mAP and True Positive Rate (TPR).
Main Methods:
- Evaluated five deep learning architectures: CenterNet, YOLOv5, Faster R-CNN, EfficientDet, and RetinaNet.
- Utilized the publicly available CBIS-DDSM mammogram dataset (1592 images).
- Trained models using L1 and L2 localization loss, assessing performance via seven key metrics.
Main Results:
- All evaluated networks achieved mean Average Precision (mAP) above 60%.
- CenterNet (Hourglass-104 backbone) achieved 70.71% mAP and 96.10% TPR.
- YOLOv5 achieved 69.36% mAP and 92.19% TPR, outperforming other state-of-the-art models.
Conclusions:
- Deep learning models, particularly CenterNet and YOLOv5, are highly effective for breast lesion detection.
- These findings suggest DL can enhance the accuracy and efficiency of breast cancer diagnosis.
- The study provides a valuable benchmark for future research in AI-driven mammography analysis.
