Mirai: Earlier Detection of Breast Cancer, Reduced Screening Harm
Mammograms are used to detect breast changes in women without signs or symptoms of breast cancer.
Health organizations worldwide support mammogram screening for early cancer detection, which has proven to reduce mortality by 20-40%.
While it's the best tool for early detection, there are many areas for improvement: false positives, false negatives, human variability in image interpretation, and lack of specialized radiologists.
Mirai, as a deep learning system, can leverage artificial intelligence to predict breast cancer formation. It includes three key innovations:
- Joint modeling of time points
- Selective use of non-image risk factors
- Ensuring consistent performance across clinical settings
This allows Mirai to provide accurate risk assessments and adapt to different clinical environments.
Mirai can predict patient risk at different future time points and incorporate clinical risk factors like age and family history (if available).
Additionally, it maintains stable predictions despite minor clinical differences (such as different mammography equipment).
A promising aspect of the model is its applicability across different races.
Mirai's accuracy is comparable for white and black women, a significant advancement given that black women have a 43% higher breast cancer mortality rate than white women.
Large-scale Validation
To integrate image-based risk models into clinical care, researchers needed to improve algorithms and conduct large-scale validation across multiple hospitals.
The research team trained Mirai on over 200,000 exams from Massachusetts General Hospital (MGH) and validated it using data from MGH, Sweden's Karolinska Institute, and Taiwan's Chang Gung Memorial Hospital.
Mirai, now installed at MGH, significantly outperforms previous methods in predicting cancer risk and identifying high-risk individuals.
It performs better than the Tyrer-Cuzick model, identifying almost twice as many future cancer diagnoses.
Moreover, Mirai maintains accuracy across different races, age groups, breast density categories, and cancer subtypes.
Adam Yala, a CSAIL doctoral student and first author of the paper, said, "Improved breast cancer risk models can enable targeted screening strategies that, compared to existing guidelines, can detect breast cancer earlier and reduce screening harms."
The team is collaborating with clinicians from various global institutions to further validate the model in different populations and study its clinical implementation.
Currently, researchers are improving Mirai by utilizing patients' complete imaging history and incorporating advanced screening technologies like tomosynthesis.
These improvements can refine risk screening guidelines, providing more sensitive screening for high-risk groups while reducing unnecessary procedures.
More Research on AI Application in Breast Cancer Detection
Beyond Mirai, Science also recommends more research on AI detection of breast cancer.
To improve breast cancer survival rates, researchers designed a wearable ultrasound device that can detect tumors in early stages, also from MIT.
Anantha Chandrakasan, Dean of MIT's School of Engineering, Vannevar Bush Professor of Electrical Engineering and Computer Science, and a co-author of the study said:
"This work will greatly advance ultrasound research and medical device design, leveraging advances in materials, low-power circuits, AI algorithms, and biomedical systems."
"It provides a fundamental capability for breast cancer detection and early diagnosis, which is key to positive outcomes."
Additionally, The New York Times previously reported on "AI Detecting Breast Cancer Missed by Doctors."
The report stated that Hungary has become a major testing ground for AI software detecting cancer, with doctors debating whether the technology will replace their medical work.
In 2016, Geoffrey Hinton, one of the world's leading AI researchers, believed the technology would surpass radiologists' skills within five years.
"I think if you work as a radiologist you're like Wile E. Coyote in the cartoon," he told The New Yorker in 2017.
"You're already over the edge of the cliff, but you haven't yet looked down. There's no ground underneath."
Hinton's words ring true. In a tweet posted by Science, one study found that doctors using AI were more likely to detect breast cancer than those not using AI.
This study also showed that AI could automatically process over half of the scans, significantly reducing radiologists' workload.
Bringing Research to Market
Science also specifically mentioned Dr. Connie Lehman on X.
Connie Lehman is a Professor of Radiology at Harvard Medical School and a radiology specialist at Massachusetts General Hospital. She is also a co-author of the groundbreaking paper mentioned at the beginning of this article.
She has been enthusiastic about the potential of computer-aided design (CAD) to improve breast cancer detection since she began working on it in 1998.