Related Experiment Video
Updated: Sep 17, 2025
07:15
Machine Learning Algorithms for Early Detection of Bone Metastases in an Experimental Rat Model
Published on: August 16, 2020
6.9K
Improving YOLO-based breast mass detection with transfer learning pretraining on the OPTIMAM Mammography Image
Pei-Shan Ho1, Hui-Yu Tsai2, Ivy Liu2
1Department of Engineering and System Science, National Tsing-Hua University, No. 101, Section 2, Kuang-Fu Road, Hsinchu, 30013, Taiwan.
Computers in Biology and Medicine
|July 2, 2025
Summary
Transfer learning with pre-trained models like YOLOv9 on the OPTIMAM Mammography Image Database (OMI-DB) significantly improves breast mass detection in mammography. This approach enhances accuracy and reduces training time for small, proprietary datasets.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Computer-Aided Diagnosis
Background:
- Mammography is crucial for early breast cancer detection, but suffers from high false positive/negative rates.
- Deep learning models for lesion detection require extensive labeled data, which is often scarce in clinical settings.
- Optimizing deep learning performance on small datasets is essential for practical application.
Purpose of the Study:
- To evaluate transfer learning, image preprocessing, and YOLOv9 for breast mass detection on small proprietary datasets.
- To compare the performance of YOLOv9 against YOLOv7 for improved detection accuracy.
- To establish guidelines for optimizing deep learning models in data-limited clinical scenarios.
Main Methods:
- Examined 133 mammography images with masses.
- Assessed preprocessing techniques like cropping and contrast enhancement.
- Investigated transfer learning using the OPTIMAM Mammography Image Database (OMI-DB) versus proprietary data training.
- Compared YOLOv9 and YOLOv7 performance.
Main Results:
- Pretraining on OMI-DB with cropped images boosted YOLOv7 performance by 13.9% mAP and 13.2% F1-score.
- YOLOv9, pretrained on OMI-DB and fine-tuned on cropped proprietary images, achieved 73.3% mAP and 76.0% F1-score.
- YOLOv9 outperformed YOLOv7 by 8.1% mAP and 9.2% F1-score in this configuration.
Conclusions:
- Transfer learning and preprocessing significantly enhance breast mass detection models.
- YOLOv9 with OMI-DB pretraining offers superior performance and reduced training time for small datasets.
- This study provides a valuable framework for optimizing deep learning in data-limited medical imaging applications.
