Improving Breast Cancer Screening with Deep Learning

Strategic Insights -

New research out of Taiwan has made significant strides in improving the detection accuracy of breast cancer through mammography. The study, led by researchers at National Tsing Hua University, employed deep learning-based computer-aided diagnosis systems to evaluate the effectiveness of transfer learning, image preprocessing techniques, and the latest You Only Look Once (YOLO) model (v9) for optimizing breast mass detection models on small proprietary datasets. The researchers' efforts aim to address the high false positive and false negative rates associated with current breast cancer screening methods.

Key Takeaways:

  • The study systematically evaluated the effectiveness of transfer learning, image preprocessing techniques, and the latest YOLOv9 model for optimizing breast mass detection models on small proprietary datasets.
  • Pretraining on the OPTIMAM Mammography Image Database (OMI-DB) dataset with cropped images significantly improved model performance, with YOLOv7 achieving a 13.9% higher mean average precision (mAP) and 13.2% higher F1-score compared to training only on proprietary data.
  • The best results were obtained using YOLOv9 pretrained on OMI-DB and fine-tuned with cropped proprietary images, yielding an mAP of 73.3% ± 16.7% and an F1-score of 76.0% ± 13.4%.
  • YOLOv9 outperformed YOLOv7 by 8.1% in mAP and 9.2% in F1-score under the condition of YOLOv9 pretrained on OMI-DB and fine-tuned with cropped proprietary images.
  • The study demonstrated that YOLOv9 with OMI-DB pretraining significantly enhances the performance of breast mass detection models while reducing training time.
  • The research provides a valuable guideline for optimizing deep learning models in data-limited clinical applications.

Statistics:

  • The study evaluated a total of 133 mammography images containing masses.
  • The YOLOv7 model achieved a 13.9% higher mean average precision (mAP) and 13.2% higher F1-score compared to training only on proprietary data.
  • The best results were obtained using YOLOv9 pretrained on OMI-DB and fine-tuned with cropped proprietary images, yielding an mAP of 73.3% ± 16.7%.
  • YOLOv9 outperformed YOLOv7 by 8.1% in mAP and 9.2% in F1-score under the condition of YOLOv9 pretrained on OMI-DB and fine-tuned with cropped proprietary images.

Sources:

  • Improving YOLO-based breast mass detection with transfer learning pretraining on the OPTIMAM Mammography Image Database. Computers in Biology and Medicine, 2025;196:110581.
  • National Tsing Hua University, No. 101, Section 2, Kuang-Fu Road, Hsinchu, 30013, Taiwan.
  • Pergamon-elsevier Science Ltd, The Boulevard, Langford Lane, Kidlington, Oxford OX5 1GB, England.