Mirai: The AI Revolution in Breast Imaging and Risk Prediction

Dr. Avisak Bhattacharjee PhD Imagine a mammogram that doesn’t just detect breast cancer but also predicts it years before it appears.That vision is now becoming reality through Mirai, an artificial intelligence (AI) model that analyses mammogram images to estimate a woman’s future breast cancer risk. Developed by researchers at MIT’s Jameel Clinic and Massachusetts General Hospital, Mirai marks a paradigm shift in breast cancer screening from reactive detection to proactive prediction. As clinicians and policymakers rethink how to make screening more personalised, risk-based, and equitable, Mirai is at the forefront of this transformation. Mirai is a deep-learning algorithm that takes a woman’s mammogram and generates an individual risk score for developing breast cancer within the next 1 to 5 years. Unlike traditional models (such as Gail or Tyrer-Cuzick) that rely heavily on questionnaires and self-reported factors, Mirai learns directly from imaging features including subtle patterns invisible to the human eye. It effectively ‘reads’ the mammogram like a radiologist with superhuman recall: 1) recognising patterns across thousands of pixels, 2) correlating tissue textures with outcomes, and 3) integrating optional clinical data (age, hormonal history, family risk). The result is a personalised, image-based forecast of breast cancer risk. Mirai was first reported in Science Translational Medicine by Adam Yala and colleagues in 2021, trained on over 200,000 screening mammograms from Massachusetts General Hospital. It was tested across multiple countries including Sweden, Taiwan, and the USA. Mirai demonstrated consistent predictive performance outperforming traditional models. Mirai’s efficacy is backed by extensive validation studies. It has been trained on about 2 million mammograms from 48 hospitals in 22 countries, demonstrating consistent predictive capability regardless of breast density, age, or equipment used. Notably, Mirai’s risk scores can more precisely flag women who are likely to develop interval cancers those that arise between regular screening visits. One of Mirai’s most impressive achievements is its performance across racial groups. Studies have shown that Mirai provides equal accuracy for Black and white women, a crucial step toward addressing breast cancer health disparities. Considering the higher mortality rates faced by Black women from breast cancer, a risk model that reliably identifies risk regardless of race is vital for equitable interventions and outcomes. One of Mirai’s most promising applications is among women with dense breast tissue, a group for whom mammography is less sensitive and cancer risk is higher. Dense tissue masks tumours on a mammogram and can make early cancers easy to miss. Mirai helps bridge this gap by learning subtle texture cues and glandular patterns associated with density, even without explicit ‘density labels’. By combining mammography data with AI, Mirai can flag women at higher risk despite a ‘normal’ report potentially guiding decisions about supplemental ultrasound, MRI, or shorter screening intervals. Mirai represents a shift towards risk-stratified breast screening, where screening frequency and modality can be tailored to an individual’s risk level. In the United States, new FDA regulations (2024) require radiology reports to notify women about breast density. Recently, Australia announced that they are also following the same trend by updating their breast density position. In contrast, New Zealand still lack national policies mandating density notification. Moreover, they have not introduced AI yet rather they are still continuing the ‘double read’ of a mammogram in their screening program by two radiologist. A recent technical review committee report added that there are significant paucity of local literature regarding breast density reporting. Mirai raises the question: could AI-based image analysis help low-resource settings stratify risk even without large radiology teams? It’s an opportunity but also a challenge as AI models need local data, governance, and cultural adaptation to avoid bias or inequity. Mirai embodies the future of breast imaging where pixels meet prediction. It invites clinicians, policymakers, and researchers to think beyond detection and towards prevention. For researchers like me studying breast density, screening policy, and equity, Mirai represents not just a technological milestone, but a call for inclusive innovation one that benefits every woman, from Washington to Wellington. @Avisak