New Study Finds Google AI Can Match or Surpass Radiologists in Detecting Breast Cancer

Improving survival rates is dependent on early and accurate detection of breast cancer, which remains a major public health concern and one of the leading causes of death among women in the United Kingdom. Historically, this critical process has been dependent on the ability of human radiologists to interpret mammograms.

A recent landmark study points to a possible paradigm change in diagnosis techniques and a significant new direction in the fight against breast cancer. This groundbreaking study strongly suggests that artificial intelligence (AI) systems can spot breast cancer faster and more precisely than human doctors.

The results imply that artificial intelligence could be a useful tool either by independently screening mammography images or by helping human experts’ work, so enabling earlier detection and greatly improved patient outcomes. With major ramifications for the future of cancer screening campaigns, this technological advancement presents a more efficient and maybe more effective screening method for millions of women.

The Biggest NHS AI Study Yet

The largest AI breast cancer screening study in NHS history was published in March 2026, with findings appearing in 2 related papers in the peer-reviewed journal Nature Cancer. The research, conducted by researchers from Imperial College London, Google, the Universities of Cambridge and Surrey, and five NHS Trusts, analyzed data from over 175,000 women. The study conducted 3 separate phases to compare Google’s artificial intelligence program with human radiologists. 

What the research actually measured

The AIMS study, Artificial Intelligence in Mammography Screening, looked at whether an artificial intelligence reader could efficiently replace the current UK standard of 2 independent human radiologists reviewing every breast cancer scan when combined with a single human reader. With an eye toward preserving the quality of care, this long-standing, decades-old 2-human-reader system served as the benchmark against which the study evaluated the AI-human pairing.

The institutions behind the work

The cooperative group of hospitals involved in the study included Imperial College Healthcare, the Royal Marsden, Royal Surrey, and St. George’s University Hospitals. Google Research provided the artificial intelligence program used in the study. The NIHR Imperial Biomedical Research Centre’s NHS AI in Health and Care Award financed this work. Researchers conducted an extensive, multi-site, multi-year project designed to depict real-world conditions inside the NHS, rather than a small-scale pilot study.

Cancer Detection Rates Shift Significantly

A study conducted between 2015 and 2016 examined 115,973 breast cancer screenings on women aged 50 to 70 across five NHS sites. The researchers found a notable increase in cancer detection through artificial intelligence (AI). When a human reader reviewed the scans alone, the cancer detection rate stood at 7.54 per 1,000 women. However, when the study included artificial intelligence as the second reader, this rate significantly increased to 9.33 per 1,000 women. Clinicians regard this rise as noteworthy because early cancer detection through improved methods can directly save more lives by enabling quicker treatment. 

Invasive cancers and first-time scans

AI identified more invasive cancers overall and performed especially well for women undergoing their first breast screening. For first-time screens, AI produced 39.3% fewer recalls than humans and delivered an 8.8% higher cancer detection rate. Fewer recalls mean fewer women facing the anxiety of an unnecessary second appointment. A higher detection rate means fewer cancers slip through unnoticed during the most critical window of early detection.

Interval cancers: The hardest ones to catch

Often, only after symptoms develop, interval cancers are those found between planned screening rounds. Usually, they are harder to treat and more hostile. 25% of these overlooked tumors were discovered by the AI tool. Evaluated in a 2025 study published in *Radiology*, a separate FDA-cleared AI algorithm correctly identified and localized 32.6% of interval breast cancers that radiologists had previously missed. The evidence repeatedly points in the same direction across several studies and instruments: artificial intelligence detects what human readers occasionally cannot.

Time Is a Resource Too

The first results of the AIMS study show that using artificial intelligence as a second reader greatly lowers the total number of scans needing human review. Specifically, compared to the previous 288,616, 32.1% fewer scans were read, translating to 195,983 readings. For the already taxed NHS system, this decrease represents a significant time saving and releases hundreds of hours annually for radiologists. The time spared from regular readings can then be directly allocated to more crucial tasks, such as biopsies, managing complex cases, and providing direct patient care.

The speed gap between AI and human readers

Monitoring 9,266 continuing cases across 12 London screening facilities, the second phase of the research found a clear structural advantage for artificial intelligence. With an average of just 17.7 minutes, the AI finished a scan analysis far faster than the first human reader, who averaged 2.08 days. For women who cannot afford delays, this dramatic speed increase is not a small improvement; it results in faster diagnosis, shortened waiting times, and the start of earlier, vital treatment.

What does the time saving mean in practice?

Professor Deborah Cunningham, a consultant radiologist at Imperial College Healthcare NHS Trust and co-author of both studies, says artificial intelligence offers a chance rather than a threat to the radiology field. She underlined that by freeing time, artificial intelligence would enable radiologists to focus their clinical knowledge on important tasks such as needle biopsies, which are essential for the cancer diagnosis process. Professor Cunningham thinks this helps to apply clinical skills more efficiently.

AI in Arbitration: A World First

In standard NHS breast screening, two radiologists independently assess each scan. If their evaluations conflict, a third reader is brought in to act as an arbitrator and provide the definitive decision. The AIMS study was the first to explore the integration of artificial intelligence into this arbitration phase, a specific clinical context not examined in previous research. Analyzing data from 50,000 women, the study demonstrated that the artificial intelligence performed comparably to human arbitrators in this role. 

The tradeoffs researchers flagged

AI flagged more uncertain cases for a third review since it generated a higher arbitration rate overall. When considering the whole process, the researchers did observe that artificial intelligence still lessened the total screening load. They also proposed that even earlier cancer detection than the two-human-reader system now permits could result from the ongoing development of the AI tool.

Why arbitration is relevant for future trials

The findings from the AIMS arbitration are particularly relevant to the forthcoming EDITH trial, according to Professor Fiona Gilbert, an academic radiologist at the University of Cambridge and an author involved in both papers.

The EDITH trial (Early Detection using Information Technology in Health) is a major international study comparing various AI tools for mammography across 30 different sites. It represents the largest study of its kind, launched with £11 million in government funding, and is expected to include approximately 700,000 women.

Significantly, the AIMS arbitration data will directly influence the EDITH trial’s protocol for managing cases where a human analysis classifies a scan as negative, but the AI technology flags it as positive.

A Workforce in Crisis

The Royal College of Radiologists published workforce census data showing a 29% shortfall of clinical radiologists in England alone, roughly 1,953 posts short of what safe patient care requires. That figure is projected to reach 39% by 2029 if no changes occur. In 2025, 36% of radiology departments reported hiring freezes, almost double the 19% recorded in 2024. Demand for CT and MRI imaging grew by 8% in 2024, while the radiology workforce grew by only 4.7%.

Breast cancer sits at the center of this crisis

Breast cancer is the most frequent cancer diagnosed in women across the UK, making up almost a third of all new cancer cases. The NHS screening program carries out over 2 million mammograms each year, leading to a substantial and increasing workload for radiologists. The existing process mandates a review of every scan by 2 specialists. This requirement, coupled with a shrinking pool of qualified staff, results in a diagnostic bottleneck, causing delays and potentially avoidable deaths.

AI as a structural solution, not just a clinical tool

The most recent review of the National Cancer Plan for England confirmed that cancer outcomes in the UK are still insufficient. According to Lord Ara Darzi, Director of the Institute of Global Health Innovation and a co-author on both AIMS papers, AI could revolutionize how the NHS approaches the prevention, detection, and treatment of diseases such as cancer. Dr. Hutan Ashrafian, also an author on both AIMS papers and based at Imperial College London’s Institute of Global Health Innovation, characterized the study’s findings as the closest AI has come to achieving a reduction in breast cancer deaths within the NHS.

What the Broader Research Landscape Shows

The AIMS study does not stand alone. A large-scale prospective German study published in *Nature Medicine* in January 2025 tracked 463,094 women across 12 sites and 119 radiologists. Radiologists using AI achieved a breast cancer detection rate 17.6% higher than those using standard double reading, equivalent to one additional breast cancer detected per 1,000 women screened. The study also found that AI-supported screening produced a similar or even lower recall rate than standard care.

South Korean and US findings reinforce the pattern

Artificial intelligence (AI) is significantly enhancing breast cancer detection rates across various studies. For example, the AI-STREAM study, a prospective multicenter study published in 2025, analyzed data from 24,543 women participating in South Korea’s national breast cancer screening program. In this study, specialist breast radiologists using AI detected 140 cancers compared to 123 without AI, representing a 13.8% increase. Detecting 120 cancers with AI against 95 without help, general radiologists observed an even more remarkable 26.4% increase.

Beyond this, Sutter Health’s AI-powered mammography program in the United States successfully raised cancer detection rates from 4.8 to over 6.0 per 1,000 screenings. Furthermore, a 2021 study by NYU researchers, published in Nature Communications, demonstrated that AI assistance not only raised radiologists’ diagnostic accuracy from 92% to 96% but also reduced the need for unnecessary biopsies.

False positives carry real human costs

AI technology significantly improves the accuracy of breast cancer screening, reducing the distress and expense associated with false positives. A false positive currently necessitates additional tests, biopsies, and weeks of patient anxiety. Research published in the Journal of the American College of Radiology demonstrated that AI systems cut false positive marks per mammogram image by as much as 69% compared to traditional computer-aided detection methods.

Furthermore, an NIH-published study indicated that AI lowered radiologist false positive rates by 37.3% without sacrificing the ability to detect actual cancer (sensitivity). This reduction in false positives directly translates to fewer unnecessary biopsies, decreased patient anxiety, and lower overall costs for the healthcare system.

What Still Needs to Happen

The second phase of the AIMS study, which involved monitoring 9,266 ongoing cases across 12 London locations, encountered an issue. During the initial 2 weeks of its live deployment, the AI’s recall rate was higher than desired, necessitating a mid-study adjustment of the criteria by the research team. Despite this change, the recall rate persisted above the established target. This highlights a crucial discrepancy: retrospective accuracy and live clinical performance are not always equivalent. This disparity demands careful consideration before proceeding with large-scale implementation.

The EDITH trial will answer the remaining questions

Specifically to answer the unresolved issues in artificial intelligence mammography, the UK government started the EDITH experiment in February 2025. The experiment tests 5 different artificial intelligence systems over 30 sites. Women will be assigned at random either 1 of 2 AI-assisted approaches or standard screening. The trial will evaluate results on all three paths; final decisions on cancer detection will rest with at least 2 human experts in every instance. The outcomes of EDITH will decide whether and how artificial intelligence (AI) enters the national routine NHS breast screening.

The Swedish MASAI trial adds confidence

Recent clinical trial data provides additional support for the use of artificial intelligence (AI) in mammography screening. The Swedish MASAI study, a randomized trial published in The Lancet in early 2026, found that AI-supported mammography was not inferior to standard double reading without AI in terms of interval cancer rate. Furthermore, the AI group demonstrated greater sensitivity (80.5% vs. 73.8%) while maintaining the same specificity. The researchers documented a measurable reduction in screen reading workload, which was an important practical finding.

The PRISM study, a clinical trial funded by a $16 million grant from the Patient-Centered Outcomes Research Institute, is currently underway at UC Davis Health and other institutions in the United States. The PRISM study, a clinical trial funded by a $16 million grant from the Patient-Centered Outcomes Research Institute, is currently underway at UC Davis Health and other institutions in the United States. This trial will evaluate AI’s clinical performance in American screening environments. This trial will evaluate AI’s clinical performance in American screening environments.

Read More: CT Scans Linked to Surprising Number of Cancer Cases, Study Finds

The Stakes of Getting This Right

Breast cancer survival rates drop sharply at later stages. The UK already has a lower proportion of breast cancers diagnosed at an early stage compared to countries including Denmark, Finland, Portugal, the United States, the Netherlands, and Sweden. The NHS screening program targets women aged 50 to 70 at three-year intervals, but uptake varies between 72 and 78% across local areas. AI that consistently detects cancers earlier, at higher rates, and with fewer false positives directly addresses one of the most critical gaps in the current system.

The NHS is running out of time to wait

According to the RCR’s February 2026 report, 100% of the clinical directors questioned voiced worries about growing imaging backlogs. About two-thirds of respondents said they lack sufficient radiologists to provide competent and safe patient care. Rising demand, slow workforce growth, and hiring freezes across more than a third of radiology departments taken together make the current situation unsustainable. Author on both AIMS papers and Google’s clinical director, Dr. Susan Thomas, said that early detection is still the most effective weapon in the fight against breast cancer and that these results really mark a turning point.

The path forward is collaborative, not replacement

Every expert quoted across the AIMS research made the same point in different words. AI is not here to replace radiologists. It is here to work alongside them. The technology takes on the volume work so that human expertise can concentrate where it matters most, in complex diagnoses, in patient-facing decisions, and in the clinical judgment calls that no algorithm can fully replicate. Professor Mike Lewis, NIHR Scientific Director for Innovation, noted that the AIMS findings demonstrate AI’s potential to improve detection, reduce unnecessary patient stress, and ease pressure on the NHS workforce simultaneously. That combination of outcomes is not common in healthcare research. When it appears, it deserves serious attention.

Read More: A new type of Artificial Intelligence can detect breast cancer 5 years before diagnosis