AI breast cancer screening finds more cancers and cuts radiologist workload, but recall risks remain. See what MASAI, AITIC, and EDITH mean for MedTech teams.
AI breast cancer screening is getting more attention. A Swedish randomised trial and a Spanish triage study show that AI can find more cancers and cut radiologist workload, while a UK study of almost 700,000 women is still running.
For pharma and HealthTech teams it’s important to establish whether the technology works and how to deploy it safely and who will pay.
Image
What Does the MASAI Trial Prove About AI Breast Cancer Screening?
The need is not in doubt. Breast cancer is the most common cancer among women, with an estimated 2.4 million women diagnosed with breast cancer and 694 000 deaths in 2024.
Many national programmes rely on two radiologists to read every screening exam. That model strains workforces as screening volumes grow, which is why AI has drawn so much attention.
Sweden’s MASAI trial is the first randomised test of AI inside a national screening programme.
It assigned more than 100,000 women to AI-supported reading or standard double reading by two radiologists. The Lancet results show that AI-supported screening caught 81% of cancers at the screening visit, against 74% in the control group.
The benefit did not stop at the screening room. Fewer cancers appeared between screening rounds, which is a key test of whether a programme works.
Earlier interim results had already shown a 44% cut in reading workload and a 29% rise in detection, without more false positives.
Key MASAI findings
The full trial reported these gains for AI-supported screening:
12% fewer interval cancers, which met the primary noninferiority aim
16% fewer invasive interval cancers, according to AJMC
27% fewer aggressive subtypes, based on the trial summary
Invasive cancer sensitivity of 78.3%, compared with 70.9% for standard reading
The authors also say that cost-effectiveness and later screening rounds need further study. The trial tested one commercial system, so results do not transfer automatically to other tools.
Regardless, the MASAI trial data is strong evidence for one workflow: AI as a support to human radiologists or other specialists. It is not a licence to deploy any algorithm anywhere.
“This is one of the riskiest industries there is because you're dealing with human lives, you're dealing with experimental protocols, and you're dealing with regulatory bodies where you might not get another shot at that clinical trial."
Image
AITIC Pushes Autonomy Further and Exposes Trade-Offs
A Spanish trial tested a bolder model.
The AITIC study, published in Nature Medicine, enrolled 31,301 women in Córdoba between March 2022 and January 2024. AI scored every exam.
Low-risk exams were treated as normal without radiologist review, and the rest were double read with AI support.
The contrast with MASAI is important. MASAI kept radiologist review on every exam and used AI to support it, while AITIC removed human review entirely for exams the algorithm judged low risk. That is a bigger step, so it carries a bigger burden of proof.
The workload gain was striking.
Radiologists read only 36.4% of exams, a 63.6% reduction. Cancer detection rose from 6.3 to 7.3 per 1,000 exams.
For health systems facing radiologist shortages, that is a powerful signal.
Why does imaging modality matter?
Results are split by technology.
In digital mammography, cancer detection improved clearly. In digital breast tomosynthesis, detection and recall stayed largely stable, while workload still fell by 65.5%. Tomosynthesis is spreading fast, so this difference will shape purchasing decisions.
The recall problem
The headline hides a catch.
In this Spanish study, the AI strategy was not noninferior to double reading on its prespecified tests, and recall rose in digital mammography.
More recalls mean more anxious patients, more follow-up imaging, and more biopsies.
A programme that saves reading time but inflates callbacks may simply move cost downstream.
Image
From Detection to Prediction and Real-World Rollout
AI breast cancer screening does not end at finding tumours.
A second wave of tools looks ahead and estimates who is likely to develop cancer in the years to come.
At the same time, large national trials are testing whether these systems work outside controlled studies.
Together, prediction and real-world evidence will decide how quickly health systems and payers adopt them.
Risk scoring from routine mammograms
Detection is only one layer.
In 2025, the U.S. Food and Drug Administration (FDA) granted De Novo authorisation to Clairity Breast, a platform that estimates five-year breast cancer risk from a routine screening mammogram.
As such, it is a prognostic tool, not a diagnostic one. It applies to women with no known diagnosis.
The developer reports validation across more than 77,000 scans from five U.S. screening centres. It also cites that up to 85% of patients have no family history, which limits traditional risk models.
The commercial logic is clear.
Risk scores could steer higher-risk women towards supplemental imaging such as MRI. They could also help sponsors identify women for prevention studies.
Pharmatica’s Insights into AI-assisted brain surgery shows the same shift from reactive to predictive care across medicine.
Why does better preventative care matter?
Earlier detection changes the patient pool that drug developers see.
If AI finds more small, early-stage tumours, the mix of cases reaching oncology clinics and trials will shift.
That could raise demand for early-stage therapies and companion diagnostics.
It may also create new questions about overdiagnosis, because finding more indolent disease does not always improve outcomes.
Sponsors, payers, and regulators will all need clearer evidence on which detected cancers truly need treatment.
Rollout evidence is still catching up
The UK’s EDITH trial will test five AI platforms across 30 sites with almost 700,000 women.
The National Institute for Health and Care Research (NIHR) is backing it with £11 million.
The aim is to let one radiologist, rather than two, complete each read. Results are still pending.
Breast Cancer Now has stressed that any change to screening must rest on robust evidence that new tools are safe and effective for women in the UK. That stance is a useful benchmark for every health system.
In the U.S., a Kaiser Permanente preprintcompared several AI vendors’ algorithms across a U.S.-representative population. It notes that screening accuracy in U.S. practice leaves room to improve.
AI in Pharma: Why the Future of Healthcare Starts With Patients, Not Tech
Kate O’Reilly, President & Chair of the Healthcare Businesswomen’s Association (HBA) Dublin-Ireland Chapter & Healthcare Transformation Partner at Roche, discusses AI in pharma, patient engagement, and the future of healthcare innovation.
Image
Questions Clinical Teams Should Ask Before Deploying AI
The evidence now supports action, but not one-size-fits-all deployment.
Before committing, decision-makers should press AI vendors and health systems on six points:
Workflow: Will AI act as a second reader, a triage tool, or an autonomous reader? MASAI and AITIC tested different models.
Modality: Does the evidence cover your mix of 2D mammography and tomosynthesis?
Recall rates: Track callbacks and biopsies, not only detection and workload.
Local validation: Ask for performance data on your own population and equipment.
Data and governance: Plan for audit trails, drift monitoring, and Electronic Health Record integration, as Pharmatica Insights on the OpenAI and Epic integration explains.
Reimbursement: Payers will want cost-effectiveness evidence, which MASAI’s researchers say is still missing.
A phased approach reduces risk.
Start with AI as a second reader, where the evidence is strongest. Monitor recall, biopsy, and interval cancer rates from day one. Expand to triage only when local data show safety across both imaging modalities.
Clinician-led testing is becoming the norm. Our look at OpenAI MentalHealthBench shows how expert panels can validate AI before it reaches patients.
Image
What Is the Bottom Line for AI Breast Cancer Screening?
AI breast cancer screening now has randomised evidence, prospective triage data, an FDA-authorised risk tool, and a national trial under way.
The strongest case is AI as a second radiologist reader or specialist, which found more cancers and cut workload in Sweden.
Autonomous triage saves more time but raises recall questions and prediction tools look promising but rest largely on developer data.
At Pharmatica, we follow the technologies, trials, and policies reshaping care, from MedTech and AI to the data behind them. Our Insights connect evidence, Regulation, and market dynamics so that teams can act with confidence when it matters most.
Pharmatica: Insight. Connection. Impact.
Frequently Asked Questions
What is AI breast cancer screening?
AI breast cancer screening uses algorithms to analyse mammograms or tomosynthesis images. The software flags suspicious areas, scores risk, or sorts exams by urgency. Radiologists then review the flagged cases. Some workflows also let AI clear low-risk exams without a human read. The goal is earlier detection, fewer missed cancers, and better use of radiologist time.
Does AI find more breast cancers than radiologists alone?
Results depend on the system, the workflow, and the population screened. In Sweden’s MASAI trial, AI-supported reading detected 81% of cancers at screening, compared with 74% for standard double reading. The AI worked alongside radiologists rather than replacing them.
Can AI replace radiologists in mammography screening?
Not yet. Most experts favour AI as a support tool with human oversight. In the AITIC trial, AI cut radiologist reading by 63.6% using AI triage, but recall rates rose in digital mammography and the strategy missed a noninferiority test.
What is the first FDA-authorised AI tool for breast cancer risk?
Clairity Breast received FDA De Novo authorisation in 2025. It predicts five-year breast cancer risk from a routine screening mammogram. It is a prognostic tool, not a diagnostic one, and does not replace standard screening or clinical judgement.
What are the risks of AI in breast cancer screening?
Key risks include higher recall rates, lower sensitivity for some in situ cancers, and results that vary by vendor and population. Cost-effectiveness data are also limited. Health systems should validate tools locally and monitor performance after launch.
Nicole (BSc Molecular Medicine, Honours Medical Biochemistry) has many years of pharmaceutical experience, having worked for top CROs and biopharma companies for more than a decade.
Did you enjoy the content?
0
Why not support
Nicole Dale
by giving this content a like
Phenotypic drug discovery is reshaping how pharma finds first-in-class drugs. This analysis covers the methods, AI integration, and strategic decisions R&D leaders face in 2026.
Learn how drug discovery scientists are evolving alongside AI, computational biology, and multidisciplinary research to accelerate therapeutic innovation.
Collaboration in drug discovery can shape pharma R&D success. Explore the evidence on pharma partnerships, knowledge sharing, governance, and innovation.
Find out how q-CAR drug discovery moves beyond static protein structures by linking protein dynamics, ligand activity, and AI to precision drug design.
FDA 2026 Draft Guidance on Alternatives to Animal Testing
The FDA's March 2026 draft guidance on alternatives to animal testing marks a pivotal shift in preclinical drug development. Here's what pharma R&D teams need to know.
Comments (0)