Artificial intelligence is beginning to reshape how breast cancer screening is delivered, and new evidence suggests its impact could be both clinically meaningful and operationally transformative.

A major trial has found that AI-supported mammography not only detects more breast cancers at an earlier stage, but also reduces the number of dangerous cancers that emerge between routine screenings. At the same time, it appears to ease the heavy workload faced by radiologists, a growing concern in many healthcare systems.

Breast cancer screening has long relied on mammography as its central tool. In many countries, women are advised to undergo screening every one or two years, depending on age and individual risk.

Yet even when guidelines are followed closely, some cancers remain undetected until symptoms appear. These so-called interval breast cancers are diagnosed after a negative screening result and before the next scheduled mammogram. They represent a persistent challenge in population-based screening.

Interval cancers are not rare. Among women screened every two years, around 10 to 20 per 10,000 will be diagnosed with an interval cancer. These cases account for roughly a quarter of all breast cancers found in regularly screened women. They also tend to be more aggressive, grow faster, spread earlier, and carry a poorer prognosis than cancers detected during routine screening. Reducing their frequency has therefore become a key goal for researchers and clinicians alike.

A large randomised trial, reported in a leading medical journal The Lancet, now suggests that artificial intelligence may offer part of the solution. The study evaluated the use of AI-supported mammography in a real-world screening setting and compared it with standard practice.

The findings indicate that AI assistance led to fewer interval cancers, improved detection of clinically relevant disease, and significant reductions in radiologist workload.

The trial, known as the Mammography Screening with Artificial Intelligence study, enrolled more than 105,000 women participating in a population-based breast cancer screening programme. Participants were randomly assigned to one of two groups. One group received standard screening, which involved independent double reading of mammograms by two radiologists. The other group underwent an AI-supported screening pathway, where an AI system analysed the mammograms and helped prioritise cases for human review.

The results were striking. Screening supported by AI was associated with a 12 per cent reduction in interval breast cancers compared with standard screening. Even more notable was the type of cancers that were prevented. The AI-supported approach resulted in fewer invasive cancers, fewer large tumours, and a substantial reduction in aggressive subtypes.

In practical terms, this suggests that AI helped identify cancers earlier, before they had time to grow and spread.

From a clinical perspective, this matters. Interval cancers are often the cases that lead to poorer outcomes, more intensive treatment, and higher mortality. Detecting these cancers earlier through screening may allow for simpler treatments and better long-term survival.

Researchers involved in the trial noted that the interval cancers that did occur in the AI-supported group tended to have more favourable characteristics, reinforcing the potential benefit.

The study also demonstrated improvements in overall screening performance. Sensitivity, the ability of a test to correctly identify cancer when it is present, increased in the AI-supported group. The sensitivity rate rose by nearly seven percentage points compared with standard screening.

At the same time, specificity, which reflects the ability to correctly identify those without cancer, remained unchanged. This balance is critical in screening, where false positives can cause anxiety, unnecessary investigations, and avoidable biopsies.

Importantly, the higher cancer detection rate did not come at the cost of more false alarms. The rate of false-positive findings remained stable, addressing one of the main concerns often raised about more intensive or technology-driven screening approaches. This suggests that AI can enhance detection without increasing harm.

The improved sensitivity was consistent across different age groups and levels of breast density. Dense breast tissue is known to make mammograms harder to interpret and is associated with a higher risk of missed cancers. The fact that AI performed well across these subgroups points to its potential usefulness in diverse screening populations.

The current findings build on earlier results from the same trial. Previous analyses showed that AI-supported mammography increased cancer detection by nearly 30 per cent compared with traditional double reading. At the same time, it reduced the number of mammograms that required full human review. In earlier reports, radiologist reading workload fell by around 44 per cent when AI was used to triage and prioritise cases.

This workload reduction has significant implications. Breast screening programmes in many countries are under strain due to workforce shortages, rising demand, and increasing imaging volumes.

Radiologists face growing pressure, which can affect job satisfaction and, potentially, diagnostic performance. Tools that safely reduce workload while maintaining or improving accuracy are therefore highly attractive.

The AI system used in the trial acts as a decision-support tool rather than a replacement for clinicians. It analyses mammographic images and assigns risk scores, helping radiologists focus their attention on the most suspicious cases.

Final decisions about recalls and diagnoses remain firmly in human hands. This model of collaboration between technology and clinicians is widely seen as a realistic and acceptable path for introducing AI into clinical practice.

Beyond detection rates and workload, the study raises important questions about cost and sustainability. Screening programmes operate within finite budgets, and new technologies must demonstrate value as well as effectiveness. While a formal cost-effectiveness analysis of the trial is still ongoing, early indications are encouraging.

Modelling studies from other European countries have suggested that AI-supported screening could be cost-effective if it leads to even modest reductions in interval cancer rates. In this trial, the observed reduction was more than double the threshold suggested by some economic models.

Fewer interval cancers may translate into lower treatment costs, fewer advanced disease cases, and reduced use of invasive procedures. When combined with savings from reduced radiologist workload, the overall economic balance may be favourable.

However, researchers caution that cost-effectiveness depends on several factors. These include the price of AI systems, integration costs, training requirements, and local healthcare structures. Long-term follow-up is also essential to confirm whether earlier detection through AI leads to sustained improvements in survival and quality of life.

Another important consideration is trust and transparency. For AI-supported screening to be accepted by both clinicians and the public, its role must be clearly communicated. Women attending screening need to understand that AI supports, rather than replaces, human expertise. Ongoing monitoring of AI performance is also essential to ensure consistent accuracy across populations and over time.

The technology evaluated in the trial has been tested in both two-dimensional and three-dimensional mammography and assessed in multiple peer-reviewed studies.

Despite the promising results, experts stress the need for caution. Screening interventions affect millions of people, most of whom are healthy. Any changes to established programmes must be backed by robust evidence. Longer follow-up is needed to assess outcomes such as breast cancer mortality, treatment intensity, and patient-reported experiences.

Several other large studies are currently underway, examining different ways of integrating AI into breast cancer screening. Their results will help determine whether the findings from this trial can be replicated in other settings and healthcare systems. Together, this growing body of evidence will inform future screening guidelines and policy decisions.

If the benefits are confirmed, AI-supported mammography could mark a significant step forward. Earlier detection of aggressive cancers, fewer interval cases, and reduced pressure on overstretched radiology services represent a compelling combination.

For patients, this could mean earlier treatment and better outcomes. For healthcare systems, it could offer a more efficient and sustainable approach to screening.

The study highlights a broader trend in medicine. Artificial intelligence is increasingly being used not as a replacement for clinicians, but as a tool to enhance human judgement. In breast cancer screening, where accuracy, efficiency, and trust are paramount, this collaborative model may prove particularly valuable.

As research continues, the focus will remain on safety, effectiveness, and equity. Ensuring that AI tools benefit all segments of the population, without widening existing disparities, will be essential.

With careful implementation, monitoring, and ongoing evaluation, AI-supported mammography has the potential to reshape breast cancer screening for the better.