Tools & Strategies News

AI Algorithms Outperform Standard Models in Cancer Prediction

New research indicates that artificial intelligence algorithms were more accurate in predicting breast cancer compared to the standard BCSC risk model.

AI for risk predictions.

Source: Getty Images

By Mark Melchionna

- Research from the Radiological Society of North America (RSNA) indicates that artificial intelligence (AI) algorithms performed better than the Breast Cancer Surveillance Consortium (BCSC) risk model in predicting the five-year risk of the disease.

Data from the Centers for Disease Control and Prevention (CDC) shows that 264,000 women and 2,400 men receive a breast cancer diagnosis annually.

Despite the various methods of predicting breast cancer, such as the BCSC risk model, their use can be taxing.

According to Vignesh A. Arasu, MD, PhD, a research scientist and practicing radiologist at Kaiser Permanente Northern California, this is mainly because the information they require can be inaccessible or difficult to obtain. However, Arasu noted that technological advances and AI could make the process of evaluating mammograms more efficient.

To compare the abilities of AI to the BCSC model, Arasu conducted a retrospective study that involved negative screening 2D mammograms from Kaiser Permanente Northern California in 2016. From a pool of 324,009 women who were deemed eligible in 2016, mammograms from 13,628 were analyzed. The study also followed the 4,584 patients from the original patient population who received a cancer diagnosis within five years.

Researchers defined three time periods based on when the diagnosis occurred: interval cancer risk, describing diagnoses between zero and one year; future cancer risk, describing diagnoses from between one and five years; and all cancer risk, encompassing the entirety of the five-year period.

Researchers used a total of five AI algorithms for the study, two of which were academic algorithms and three of which were commercially available. After comparing their performance to the abilities of the BCSC risk model, researchers found that the AI algorithms performed better than the standard risk model.

“All five AI algorithms performed better than the BCSC risk model for predicting breast cancer risk at 0 to 5 years,” said Arasu in a press release. “This strong predictive performance over the five-year period suggests AI is identifying both missed cancers and breast tissue features that help predict future cancer development. Something in mammograms allows us to track breast cancer risk. This is the ‘black box’ of AI.”

Beyond this, the AI algorithms presented several other benefits. Researchers also noted that certain AI algorithms performed well in predicting those at risk of interval cancer. This is critical, as this generally requires follow-up mammogram readings.

Also, even AI algorithms that did not have a long training duration performed well.

“We’re looking for an accurate, efficient and scalable means of understanding a women’s breast cancer risk,” said Arasu. “Mammography-based AI risk models provide practical advantages over traditional clinical risk models because they use a single data source: the mammogram itself.”

AI is playing an increasingly significant role in cancer prediction and detection, serving as the foundation of many research efforts.

A large grant from the National Cancer Institute in November 2022 led researchers from the University of California Davis to fuel AI projects to enhance breast cancer screening and risk prediction. Through these efforts, researchers aimed to reduce health disparities.

Often, certain types of regular screening can lead to false positive results. The grant, however, will support researchers as they test whether new AI and imaging features can improve risk prediction models.