Breast cancer is still the leading cause of cancer death in women in 110 countries. Despite decades of research and public awareness, millions of women are still diagnosed too late. Now, artificial intelligence may offer a way forward.
Breast cancer remains the leading cause of cancer death in women in 110 countries, with 2.3 million diagnoses and 685.000 deaths recorded globally each year.
Screening plays a vital role in catching the disease early, improving outcomes, and ultimately saving lives. Despite decades of public attention and scientific advancement, far too many women are still diagnosed after the disease has already taken hold.
Worldwide, radiologists can achieve remarkable accuracy in detecting breast cancer, although results vary depending on national policies, infrastructure, and access to care. Yet, even in high-resource settings, challenges persist. Radiologists must interpret hundreds of mammograms each week–most of which are regular–and their performance can vary with experience, fatigue, and training. Moreover, depending on the healthcare system, a diagnosis might involve multiple repeated steps. European screening programmes, for instance, involve two radiologists to read each mammogram. If at least one deems the case as suspicious, then a consensus conference is held with at least the two initial readers and one head radiologist. If the suspicious finding persists, the woman is recalled for further diagnostic assessments. But the most critical cases happen in the so-called healthcare deserts, where access to skilled radiologists is limited or nonexistent.
These challenges underscore the urgent need to rethink how we screen and diagnose breast cancer, and recent developments suggest that we may be entering a transformative new era.
In 2025, the U.S. Food and Drug Administration (FDA) granted De Novo authorisation to Clairity Breast, the first AI-powered platform capable of predicting a woman’s risk of developing breast cancer over the next five years, based solely on a standard mammogram.
This marks a significant milestone in the history of Computer-Aided Detection (CAD), which began in the 1990s with promise but limited results. Early CAD systems failed to significantly improve real-world diagnostic performance due to hardware and software constraints, but the rise of deep learning in the 2000s changed everything.
In 2000, just 12 scientific papers were published on AI-assisted breast cancer detection. By 2020, that number had grown to 456– averaging 1,3 papers per day–reflecting the growing belief in AI’s potential to assist clinicians in high-stakes diagnostic tasks.
With the availability of better AI models, then came real-world applications and one of the most compelling validations was realised in Sweden in 2023. The MASAI study (MAmmography Screening with Artificial Intelligence) enrolled 80.000 women aged 40 to 74 across four cities: Malmö, Lund, Landskrona, and Trelleborg. Participants were randomly assigned to standard double reading by radiologists or to a screening pathway that included AI-supported reading.
The AI system used in the study had been trained on more than 200.000 mammograms from ten countries, ensuring it reflected diverse populations and equipment types. The results were striking: the AI-assisted group detected 20% more cancers than the standard group, while reducing radiologist workload by 44,3%.
The momentum continued in Germany in 2025 with the PRAIM study (PRospective multicenter observation Study of an integrated AI system with live Monitoring). Embedded within the national screening program, PRAIM involved over 460.000 women, 119 radiologists, and 12 screening sites using equipment from 5 different vendors.
Radiologists were free to choose whether or not to use AI support. If at least one of the two readers used AI, the case was marked as “AI-supported”. Otherwise, it was part of the control group with the standard two-readers procedure. The findings were consistent with the MASAI study: AI-supported readings detected 17,6% more cancers and reduced radiologist workload by more than 50%, allowing more time to focus on complex cases or patient interaction.
But effective screening isn’t just for clinicians. It’s also about patients and trust. A U.S.-based survey of 518 patients found that 71% were comfortable with AI being used as a second reader, but fewer than 5% supported AI making decisions alone. Preferences varied across different social groups but reinforced a broader conclusion: patients want AI to augment–not replace–expert judgement.
And then we come to the present day, with the FDA granting De Novo authorisation to Clairity Breast. This platform, developed with the support of the Breast Cancer Research Foundation (BCRF), leverages AI to detect subtle patterns in breast tissue that might signal future cancer development, even if the mammogram appears normal to the human eye.
Unlike traditional risk models, which rely on age, family history, or questionnaires, Clairity Breast uses the mammogram itself as the primary source of insight. It was trained on millions of images and validated across 77.000 mammograms from five screening centers serving diverse populations.
Patients identified as high risk can be referred for supplemental imaging such as MRI, more frequent screening, or preventive strategies like medication, lifestyle changes, or genetic counselling. Conversely, those at lower risk may avoid unnecessary testing and the anxiety that comes with it.
While Clairity Breast may not be widely accessible yet, its FDA authorisation opens the door to a new frontier in cancer screening, where AI can act as a partner in prediction and prevention.
And once that door opens, there’s no turning back.
References:
