Related Experiment Video
Updated: Nov 19, 2025
09:36
Changes in Mammary Gland Morphology and Breast Cancer Risk in Rats
Published on: October 16, 2010
30.0K
Toward robust mammography-based models for breast cancer risk
Adam Yala1,2, Peter G Mikhael3,2, Fredrik Strand4,5
1Department of Electrical Engineering and Computer Science, Massachusetts Institute of Technology, Cambridge, MA 02139, USA. adamyala@csail.mit.edu.
Science Translational Medicine
|January 28, 2021
Summary
A new deep learning model, Mirai, significantly improves breast cancer risk prediction using mammograms. This advanced model identifies high-risk patients more accurately than existing methods, paving the way for earlier detection and personalized screening strategies.
Area of Science:
- Artificial Intelligence in Medical Imaging
- Oncology
- Radiology
Background:
- Accurate breast cancer risk assessment is crucial for effective screening and early detection.
- Current risk models have limitations in accuracy and broad applicability.
- Deep learning models show promise but require further refinement for clinical integration.
Purpose of the Study:
- To develop and validate Mirai, a novel deep learning model for predicting breast cancer risk using mammography.
- To enhance the accuracy and consistency of risk predictions across diverse populations and imaging equipment.
- To compare Mirai's performance against established risk assessment models.
Main Methods:
- Developed Mirai, a mammography-based deep learning model trained on a large US dataset (MGH).
- Validated Mirai on independent datasets from Sweden (Karolinska) and Taiwan (CGMH).
- Compared Mirai's 5-year risk prediction performance (ROC AUC) against the Tyrer-Cuzick model and other deep learning models (Hybrid DL, Image-Only DL).
Main Results:
- Mirai achieved high C-indices across all test sets (0.76-0.81).
- Mirai demonstrated significantly superior 5-year ROC AUCs compared to Tyrer-Cuzick, Hybrid DL, and Image-Only DL models (p < 0.001).
- Mirai identified a higher proportion of future cancer cases as high-risk compared to existing methods.
Conclusions:
- Mirai represents a significant advancement in mammography-based deep learning for breast cancer risk prediction.
- The model's performance across diverse datasets suggests potential for broad clinical utility.
- Mirai offers improved accuracy in identifying high-risk individuals, supporting targeted screening and potentially reducing screening-related harms.
