Insight
Changes in AI mammogram risk scores help predict future breast cancer
Changes in AI mammogram scores may help predict breast cancer years before diagnosis, research involving more than 54,000 women suggests.
Scores rose steadily among women who later developed the disease but remained broadly stable among those who did not.
The increase could be detected up to six years before diagnosis and became much steeper during the final two years.
Researchers led by Professor Constance Lehman, of Harvard Medical School and healthcare technology company Clairity, analysed screening mammograms taken between 2009 and 2019.
They used a validated, open-source deep learning model to calculate five-year breast cancer risk scores from the images alone.
Deep learning is a form of artificial intelligence trained to recognise complex patterns in large amounts of data.
The model examined the whole mammogram rather than relying on a limited, predetermined feature such as breast density.
Models of this kind have performed better than traditional risk models and breast density alone when estimating a woman’s five-year breast cancer risk.
The study initially included 239,703 consecutive two-dimensional screening mammograms from 89,882 patients across six imaging sites spanning urban tertiary, community-based and rural settings.
All were standard bilateral full-field digital mammography examinations, taken with or without digital breast tomosynthesis.
Digital breast tomosynthesis uses multiple low-dose X-ray images to create a three-dimensional view of the breast.
After exclusions, the final analysis involved 54,014 women with a median age of 61 and a total of 158,807 mammograms.
Each woman contributed one index examination and up to six previous annual mammograms. Women had a median of three scans each.
For women who developed cancer, the index examination was their final screening mammogram within the year before diagnosis. For the cancer-free group, it was their final mammogram during the five-year study period.
The model did not use demographic information, clinical records or historical imaging data when calculating each score.
Of the women included, 817, or one per cent, were diagnosed with breast cancer within 365 days of their index examination.
This included 451 women, or 55 per cent, with invasive breast cancer and 118, or 14 per cent, with ductal carcinoma in situ, known as DCIS.
DCIS occurs when abnormal cells are found inside a milk duct but have not spread into the surrounding breast tissue.
The cancer type was unknown for the remaining 248 patients, representing 30 per cent of the cancer group.
A total of 682 cancers, or 83 per cent, were detected through screening, while 135, or 17 per cent, were interval cancers diagnosed between routine mammograms.
The other 53,197 women were not diagnosed with breast cancer during follow-up and formed the cancer-free comparison group.
Professor Lehman said: “We observed clinically relevant differences in risk trajectories between women who did and did not develop cancer. The increase in scores among cancer patients was detectable as early as six years prior to diagnosis and became more pronounced over time.”
Among women later diagnosed with the disease, the median score rose from 2.1 five to six years before diagnosis to 6.6 at the index examination.
Scores among cancer-free women remained stable, with median values ranging from 1.8 to 2.2 throughout the study.
The rise among women who developed cancer was steepest during the two years before their index examination.
Professor Lehman said: “These findings demonstrate signals, invisible to the human eye, in the image alone can predict future risk. This is exciting, because 85 per cent of women diagnosed with breast cancer do not have a significant family history of breast cancer or known genetic mutations.”
Most breast cancers are considered sporadic, meaning they are not driven by inherited genetic changes or a family history of the disease.
Traditional risk models have a limited ability to distinguish between women who will and will not develop breast cancer when used across large screening populations.
Researchers said tracking how scores change over time could provide more information than calculating risk at a single appointment.
Professor Lehman said: “AI-derived risk scores can identify patients who are otherwise predisposed to the disease, and our findings demonstrate that image-based AI risk scores evolve over time and that changes in those scores may provide additional information about future breast cancer risk.”
The patterns remained consistent when women were grouped by age and breast density.
Breast density describes the amount of fibrous and glandular tissue visible on a mammogram. Dense tissue can make cancers harder to detect and is also associated with an increased risk of the disease.
Researchers said image-based scores could support personalised screening and risk-reduction strategies without relying on self-reported or inconsistent clinical information.
Professor Lehman said: “These trends remained robust across subgroups defined by age and breast density, further supporting the generalisability of our findings. This is particularly relevant given persistent disparities in screening performance across patient populations. A dynamic biomarker approach grounded in the imaging data could mitigate some of these disparities by enabling risk-based personalisation that does not rely on self-reported or inconsistent clinical data.”
A biomarker is a measurable sign that can indicate a person’s health, disease risk or response to treatment.
Changing scores could eventually help clinicians identify women who may benefit from additional imaging or measures intended to reduce their risk.
Professor Lehman said: “With the power of AI, computer vision, and the ability to extract predictive data, we are able to apply the power of imaging to risk assessment and preventing disease from developing. Having a dynamic risk score opens up a whole new domain of more effective preventive therapies for breast cancer, similar to how we screen for and treat patients with high cholesterol and hypertension.”
AI image-based risk scores are included in the 2026 National Comprehensive Cancer Network guidelines.
The guidelines recommend that, from the age of 35, women with an elevated five-year risk score of more than 1.7 per cent consider breast MRI alongside annual mammography.
An AI image-based model approved by the US Food and Drug Administration is already being used to calculate five-year breast cancer risk at selected US healthcare institutions.
Black women in the UK want clearer, more accessible breast cancer screening information, research has found.
The study looked at why Black African and Black Caribbean women are less likely than white women to attend breast screening.
Researchers at the University of Surrey held focus groups and interviews with 47 Black African and Black Caribbean women aged 50 to 71.
Women in this age group are routinely invited for NHS breast screening.
The researchers said only 45 per cent of Black women attend screening, compared with 63 per cent of white women.
Anietie Aliu, lead author, postgraduate researcher at the University of Surrey and registered nurse, said: “Diagnosing breast cancer early can dramatically improve a person’s chance of survival.
“Breast cancer screening plays an important role in this by identifying the cancer and ensuring a person receives speedy treatment.
“Despite the importance of screening, Black women are less likely to attend appointments than white females.
“This puts them at risk of a potential cancer being diagnosed late and spreading to other areas of the body. We need to understand what is preventing Black women from attending these appointments and help identify ways to remove such barriers.”
The study found a need to increase awareness of breast cancer screening, especially among women less familiar with the service.
Some women, particularly those born outside the UK, knew little about breast screening before receiving their first invitation.
Others questioned why they needed screening when they had no symptoms.
The importance of trusted conversations was also identified.
Researchers found that some Black women expected their GPs to speak to them about breast screening, particularly before they reached screening age.
Although NHS breast screening is organised through national screening services, researchers said GPs often have established relationships with patients and may be well placed to offer brief advice on preventive care, including breast screening.
Participants called for stronger links between GP practices, breast screening services and Black community champions.
They said this could help women receive trusted information, ask questions and feel reassured.
Faith and religious beliefs also shaped decisions for some women.
Some Black African Christian women said illness, including cancer, was not permitted by God in their bodies, while others saw screening as a personal choice that did not conflict with Christian faith.
Muslim women highlighted the importance of being able to state their religion on medical appointment forms to help ensure they were seen by a female mammographer.
A mammographer is a healthcare professional trained to carry out breast screening scans.
Aliu added: “Breast screening can save lives, but our findings show that attendance is shaped by multiple factors, not just awareness, although awareness remains important.
“Women need relatable screening information, reassurance, flexible appointments and services that are accessible within their communities.
“Many felt that invitation letters were too formal, and that leaflets and media imagery did not reflect them, making it harder to relate to screening.”
Dr Afrodita Marcu, senior research fellow at the University of Surrey and member of the research team, said: “We need a more collaborative approach, where primary care, screening services and community voices work together to support women before, during and after the invitation.”
The researchers said future breast screening interventions should be designed with Black women, rather than for them.
They said user-friendly and culturally relevant resources, developed with communities, healthcare professionals and screening services, could improve understanding, reduce fear and make breast screening feel more accessible and reassuring.
Dr Robert Kerrison, associate professor of cancer care at the University of Surrey, said: “There is no question that breast screening can be lifesaving, but we need to make it easier for women to understand, access and feel reassured by the programme.
“This means improving communication, addressing practical barriers and making sure healthcare professionals and community partners are supported to provide clear and trusted information.”
The team has also explored healthcare professionals’ perspectives and worked with stakeholders to develop user-friendly materials with Black women.
Researchers said this co-designed approach could help ensure breast screening messages are culturally relevant, practical and shaped by the people they are intended to support.
News
“Women’s voices should be heard and pain should never be ignored,” says Wales’s first Women’s Health Minister
Women’s pain should not simply be endured, Wales’s first women’s health minister has said.
Delyth Jewell said she was determined to tackle the normalisation of pain in women’s healthcare and ensure women’s voices are listened to.
Speaking during a Women’s Health Summit at the Temple of Peace on Thursday, July 16, she said: “For too long, women’s health has been treated as an afterthought. No woman should be afraid to speak up about pain or things that don’t feel right.
“Women should be believed about their bodies, and I am determined to change the culture that has let too many women down.”
She added: “Women’s voices helped create the Women’s Health Plan. Now we’re making sure those voices continue to shape what comes next.”
The summit brought together clinicians, researchers and women with lived experience to tackle the normalisation of pain in healthcare and identify how women’s voices can better shape NHS services.
Lived experience means insight from people who have personally gone through a health issue or used healthcare services.
The event focused on pain linked to clinical procedures and long-term health conditions, drawing on research evidence, clinical expertise and women’s personal experiences.
Following the summit, minimum standards for service user engagement will be drafted to ensure women’s voices continue to influence the delivery and future priorities of the Women’s Health Plan.
Service user engagement means involving people who use health services in decisions about how care is designed, delivered and improved.
Work will also begin to refresh and strengthen the plan, including gathering feedback directly from women across Wales.
The NHS Wales Women’s Health Plan was developed after discovery work in 2022, when women across Wales shared their experiences of healthcare.
Many said they had not felt listened to, had symptoms dismissed or had lived with pain for years before receiving a diagnosis.
Hosted by Amanda Ducach, CEO, and Morgan Rose, chief science officer, EmaEQ
Healthcare companies have spent the last two years hearing the same advice: get AI into your product. Few have been told what that actually takes.
Most default to the fastest option. Plug in a general-purpose model, wrap it in a chat window, and call the box checked. It looks like progress on a roadmap slide. It rarely holds up once a real patient is on the other end of it.
We’ve spent years building AI specifically for healthcare, and the lesson that keeps repeating itself is simple: accuracy is not the same thing as trust, and trust isn’t something you bolt on after launch. It has to be part of how the system is built from the first line of code, not a feature added once regulators or users start asking questions.
That distinction is the whole reason clinical accuracy gets treated as a checkbox instead of a discipline. A model can sound confident and still be wrong in ways that matter enormously in a health context.
Knowing the difference, and building for it deliberately, is what separates AI that’s genuinely safe for care from AI that’s simply fast to ship.
On July 20th, we’re hosting a live conversation about exactly this: what companies should be paying attention to before they choose an AI to build with, what clinical accuracy really requires, and the pillars we hold every AI system to before it gets anywhere near a patient’s care.
The Healthcare AI Playbook Webinar: July 20th, 1:30-2pm EST, live on LinkedIn.
Register here: https://www.linkedin.com/events/7482643171823509504?viewAsMember=true
If your team is building anywhere near healthcare, or evaluating what’s already in your product, this is the conversation we think the industry needs right now.
Diagnosis2 weeks agoTwo “gamechanger” tests set to speed up endometriosis diagnosis on the NHS
Insight2 weeks agoWomen with PMOS should have annual NHS checks, new guidance says
News5 days agoNew menopause drug approved for use by NHS in Scotland
Cancer2 weeks agoThousands of women could avoid painful cancer exam with new AI blood test
Entrepreneur2 weeks agoApplications open for the third W Accelerate with Merck KGaA and M Ventures
Fertility2 weeks agoOlder women face lower chance of fertility treatment working, even with donor eggs, study finds
News2 weeks agoWe built Ema like a nurse: Here’s why that matters
News2 weeks agoExperts call for regulation of toxic period product chemicals
3 Comments