AI that predicts breast and lung cancer years ahead: what it means for you

Your routine mammogram or lung CT holds more information than the eye can see. New AI tools read it to estimate your risk over the next five or six years.

Why traditional risk tools miss so many women

Most breast cancer risk calculators used today, such as Gail or Tyrer-Cuzick, are built on questionnaires: age, family history, age at first period, previous biopsies and genetic results. The only imaging input is breast density, judged by eye into four broad categories.

The problem is that most women who develop breast cancer have no family history at all, so questionnaire-based tools flag relatively few of them in advance. Lung screening has a similar gap: it is offered mainly to older heavy smokers, while lung cancer in people who never smoked is rising.

How AI reads risk from a normal scan

A radiologist looks for something visible: a mass, calcifications, a distorted area. A predictive AI model does something different. It analyses the texture and structure of the whole breast or lung, pixel by pixel, and compares it with hundreds of thousands of earlier scans from women and men whose outcomes are known.

Two patterns matter most. The first is subtle tissue texture that the human eye cannot separate from normal. The second, in breast imaging, is small differences between the left and right breast. Healthy breasts are broadly symmetrical, and researchers at Duke University showed that local asymmetry is one of the main signals the models use.

The output is a number: your estimated chance of being diagnosed within a set period, typically one to five years for breast and up to six years for lung.

The main tools, and where they stand

Clairity Breast

Developed by a team led by Dr Connie Lehman (Harvard and Massachusetts General Hospital), Clairity uses only a routine 2D screening mammogram. It needs no questionnaire or genetic test. In May 2025 the FDA granted it De Novo authorisation, which created a brand-new device category: software that predicts future breast cancer risk from screening images.

Mirai and AsymMirai

Mirai, built at MIT and Massachusetts General Hospital, uses all four standard mammogram views and gives a risk estimate for each of the next five years. It has been tested in hospitals in the US, Sweden and Taiwan. AsymMirai, from Duke, is a simpler version that shows doctors which area of the breast drives the score. Both remain research tools.

Sybil (lung)

Sybil, also from MIT and Massachusetts General Hospital, reads a single low-dose chest CT and predicts lung cancer risk for up to six years, without needing earlier scans or clinical details. It was validated on patients from the US National Lung Screening Trial, Mass General and a Taiwanese hospital where many patients had never smoked.

Optellum Virtual Nodule Clinic (lung nodules)

Optellum works differently. It does not predict long-term risk; it scores how likely an already-found lung nodule is to be cancer, helping doctors decide between biopsy and watchful waiting. It is FDA-cleared and used in NHS hospitals in England.

How accurate are they?

Accuracy is usually reported as AUC, where 0.5 is a coin toss and 1.0 is perfect. Questionnaire-based breast models typically score around 0.60 to 0.62.

• Mirai: about 0.84 for predicting cancer within one year and 0.71 to 0.75 at five years, across several countries and ethnic groups.
• Sybil: 0.86 to 0.94 at one year and 0.75 to 0.81 at six years across three cohorts (Journal of Clinical Oncology, 2023).
• Clairity: trained on more than 400,000 mammograms. Research presented at RSNA in late 2025 showed it separated five-year risk better than breast density, including in women under 50.

What changed in 2026

On 14 April 2026 the National Comprehensive Cancer Network (NCCN), whose guidelines shape US cancer care, added image-only AI risk assessment to its breast cancer screening guidelines. A five-year AI risk of 1.7% or higher is now treated as increased risk, and the assessment can start from age 35. For women above the threshold, doctors may discuss extra imaging such as MRI, more frequent checks, genetic counselling or risk-reducing options.

In October 2026 the digital health company Everlywell began selling the Clairity assessment directly to US women aged 35 and over who have had a 2D screening mammogram in the past 12 months. The woman uploads her mammogram files, a licensed physician reviews the result, and the price is $249.

Availability around the world

• United States: Clairity is used in some hospitals, starting with Beth Israel Deaconess Medical Center in Boston, and is sold direct-to-consumer. US insurers generally treat it as not yet covered, so most people pay themselves.
• United Kingdom: Optellum is used in NHS lung pathways. We found no official data on NHS use of Clairity or Mirai.
• European Union: ProFound AI Risk is CE marked and offered by some private radiology networks.
• Turkey: AI software sold as a medical device must hold an EU CE certificate and be registered in the national product tracking system (ÜTS). We found no official data on Clairity, Mirai or Sybil being offered in Turkey.
• Research access: Mirai and Sybil are mainly available through clinical trials at US academic centres.

Risks and limits

• Extra tests and anxiety. A high score usually leads to more imaging, such as MRI, and sometimes biopsies that turn out benign. In one US hospital study, about a third of women in the top risk group were called back for more tests.
• Overdiagnosis. Earlier detection can uncover slow-growing changes that would never have caused harm.
• Not equally accurate for everyone. Simpler models performed less well in some groups, such as Black women and women aged 50 to 70 in the AsymMirai study. Scanner brand can also affect results, which is why the FDA requires testing across different machines.
• A black box. For most models, a radiologist cannot point to a visible finding that explains a high score.
• Not for everyone yet. Clairity is indicated for women aged 35 and over without a personal history of breast cancer or symptoms. It is not validated for men or after mastectomy.

Trials to watch

• MIRAI-MRI (NCT05968157): offers breast MRI to women whose Mirai score is in the top few percent, to measure how many extra cancers are found.
• RESOLVE: a Mass General study, funded in 2026, testing whether Sybil can safely cut unnecessary follow-up scans and biopsies for lung nodules.
• WISDOM: a large US trial of risk-based breast screening intervals, now adding AI image scores to genetic and lifestyle risk.

What to ask your doctor

• Does my imaging centre offer an AI risk score, and is it an authorised product or a research tool?
• If my score is above 1.7%, what extra screening would you recommend, and who pays for it?
• How does the AI score fit with my family history or genetic results?
• For lung screening: am I eligible, and is AI used to assess any nodules found?

Questions patients ask