New AI tool predicts breast cancer recurrence faster

New artificial intelligence tools could soon change the way doctors understand breast cancer risk, providing faster answers and more precise guidance at critical moments in treatment. The technology could help determine whether a patient’s cancer is likely to come back, an issue that affects treatment decisions for millions of people.

Researchers at New York University developed the test by combining digital images of tumor tissue with basic clinical information. Their findings show that the system can match or outperform widely used genetic tests.

Breast cancer treatment has advanced in recent decades. Today, many patients survive and continue to live fulfilling lives. Still, recurrence remains a serious concern.

Physicians often face difficult choices when deciding how aggressively to treat a disease. Some patients receive unnecessary chemotherapy. You may need more treatment than you are receiving.

“Breast cancer is not a single disease, and deciding how aggressively to treat it is often difficult,” said Krzysztof J. Geras, who led the study.

To make these decisions, clinicians often rely on genomic testing. These tests analyze gene activity in tumor tissue to estimate the risk of recurrence. Although convenient, it also has its limitations.

Results may take several weeks. It is also expensive and requires specialized processing that uses up valuable tissue samples.

The new approach is based on what doctors already use routinely. When cancer is diagnosed, a tissue sample is placed on a glass slide and examined under a microscope.

The AI ​​system analyzes these same slides in digital format. Look for patterns invisible to the human eye.

Those findings are then combined with clinical details such as tumor stage, patient age, and hormone receptor status. The result is a score that estimates the likelihood that the cancer will come back.

“This study shows that an AI test can read the same tumor slides that pathologists are already examining and, in combination with basic clinical details, accurately estimate the likelihood that a patient’s cancer will come back,” Geras said.

This system uses a technique called self-supervised learning. Rather than relying solely on labeled examples, it first studies large amounts of data to learn its own patterns.

“Model accuracy doesn’t just come from hand-labeled data,” says Yann LeCun. “This comes from self-supervised pre-training that first learns rich representations, which then translates into strong downstream performance. This recipe should generalize far beyond breast cancer, and more broadly, it’s the kind of new AI science that these difficult problems demand.”

This approach allows the model to detect subtle features that would otherwise go unnoticed. It also makes it more adaptable to different types of data.

To evaluate the system, researchers analyzed data from more than 3,500 patients. These patients came from 15 groups from 7 countries.

The model showed good performance in classifying patients into high- and low-risk categories. Standard statistical measures were used to check its accuracy.

One key metric, known as the concordance index, showed that the system reliably ranked patients by risk. Another measure, hazard ratio, identified clear differences between risk groups.

The AI ​​test also showed positive results in different types of breast cancer. This includes triple-negative and HER2-positive cases where existing genomic tests are often inadequate.

This wide range of performance suggests that this system has the potential to help a wider range of patients.

One of the most immediate benefits is speed. Traditional genomic testing can take several weeks to complete. The AI ​​system provides results within hours once the slides are digitized.

This quick response can reduce stress for patients waiting for answers. It may also help doctors decide on treatments more quickly.

Cost is another advantage. This system uses existing slides, eliminating the need for additional processing in the laboratory. This can potentially make testing more accessible.

Also save tissue samples. Unlike genomic testing, which consumes a portion of the sample, AI approaches leave behind material that can be used in the future.

The system also provides clearer risk categories. Current genomic tests often classify patients into low-, intermediate-, or high-risk groups. Intermediate groups can be difficult to interpret.

AI models provide a clearer separation. This may help doctors make more confident decisions about treatment.

For patients, this could mean avoiding unnecessary chemotherapy or receiving more targeted treatment if needed.

The system also showed excellent results over various time periods. We were able to predict outcomes several years later, including the risk of distant recurrence.

The study highlights broader changes in medicine. Rather than relying on one type of data, researchers are combining multiple sources to better understand diseases.

By integrating imaging and clinical information, AI testing provides a more complete picture of each patient.

It also shows how artificial intelligence can enhance rather than replace existing medical tools.

Despite the promising results, researchers caution that more research is needed. This system needs to be tested in randomized clinical trials to confirm its value in real-world decision-making.

They also aim to investigate how the model can predict response to specific treatments. This may further improve its usefulness in guiding care.

Still, the findings represent a meaningful advance. They suggest that faster, more accessible tools could soon help doctors and patients make better decisions.

This research could improve how breast cancer is treated by providing a faster and more accurate risk assessment. Patients can receive more personalized care based on their personal risk rather than general guidelines.

Lower costs and faster results could make advanced tests available to more people. This is especially important in regions where access to genomic testing is limited.

Preserving tissue samples opens up new possibilities. Doctors may be able to perform additional tests later, which may support new treatments or research.

In the long term, this approach could extend beyond breast cancer. Similar systems could be developed for other diseases and help improve diagnosis and treatment across medicine.

By combining speed, accuracy, and accessibility, this technology provides a path to more informed and effective care.