LUND, Sweden — When 54-year-old Helena Larsson learned that a routine mammogram had caught an early breast tumour, her first question wasn’t about radiation or chemo. It was about her armpit.
“Will you have to take out my lymph nodes?”
For decades, the answer has been almost automatic: yes. Surgeons remove one to three “sentinel” nodes to see if cancer has spread. If they find malignant cells, more nodes come out. The operation is considered minor, but its side effects—numbness, swelling, and chronic pain—can linger for years.
Researchers at Lund University may have found a way to avoid it. Their new AI system can read standard mammograms and predict, with striking accuracy, whose lymph nodes are actually at risk. The study, presented last month at the European Breast Cancer Conference in Barcelona, suggests that four in ten patients could safely skip node surgery altogether.
Teaching a Computer to Read Between the Pixels
Pathologist Johan Hartman and data scientist Linda Ahlgren trained a deep-learning model on more than 6,000 digitized screening mammograms, paired with details like age, tumour size, grade, and hormone-receptor status.
The AI learned to detect faint cues—tiny shifts in calcification, tissue density, and blood-vessel structure—that even experienced radiologists can’t see, but that correlate with microscopic cancer spread.
When tested on 1,265 Swedish women diagnosed between 2009 and 2017, the model flagged 41.7 percent as having less than a 5 percent chance of lymph-node involvement. Among those “low-risk” cases, the AI missed a true positive just 3.2 percent of the time—well within the safety limits international oncology guidelines allow.
Why It Matters
Each year, about 2.3 million women worldwide are diagnosed with breast cancer. For most, a sentinel-node biopsy is standard. Yet in many cases, the surgery doesn’t change treatment at all.
“Every avoided biopsy means one less general anaesthetic, one less scar, and one less chance of lifelong lymph-oedema,” says oncologist Lisa Kurland, who was not involved in the study. “If these results hold up, we could spare hundreds of thousands of women unnecessary operations every year.”
The Road from Lab to Clinic
The Lund team is cautious. Their AI has only been tested on early-stage, screen-detected cancers from Sweden—a country with robust screening programs and relatively uniform imaging standards.
“We need to validate this in places like the UK, US, and Asia—settings with different demographics and mammography protocols—before changing practice,” says Hartman.
A prospective clinical trial, in which low-risk patients skip the biopsy and are followed for five years, is set to begin in 2026.
Still, change may come sooner. Skåne University Hospital in Malmö plans to pilot the system this winter, offering selected low-risk patients’ ultrasound-only nodal assessment.
“If a patient understands the residual risk and chooses to avoid surgery, I think that’s reasonable,” says breast surgeon Emma Böhler.
What Women Can Ask Now
Until large trials confirm safety, most patients will still undergo node biopsies. But they can start asking sharper questions:
- What’s my personal risk of lymph-node spread based on imaging and tumour biology?
- Does this clinic use any AI decision-support tools?
- If my risk is very low, are there monitoring or surveillance options instead of surgery?
For Helena Larsson, who took part in the Lund study, the AI brought relief. Her scan was labelled low-risk. She chose a lumpectomy and radiation, skipping the node operation.
Six months later, her follow-up scans are clear. “It feels almost too good to be true,” she says, flexing her left arm, pain-free. “But if a computer can look at my mammogram and say, ‘Relax, your nodes are fine,’ maybe it’s time we listen.”