Breast cancer screenings can save lives through earlier detection. Mammograms are among the most common tests, although MRI scans and ultrasounds are other possibilities. Numerous studies show artificial intelligence (AI) could improve outcomes in various ways. Such results enable health care providers to support their patients and have more potential interventions to explore.
Reducing Technician Workloads
A common method of reducing false positives during breast cancer screenings is to have two radiologists examine a patient’s results. Agreement from both minimizes the chances of misinterpretation. However, some health care systems face radiologist shortages, increasing these professionals’ work-related burdens.
A Swedish trial aimed to tackle this challenge by assigning women to receive either standard or AI-assisted breast cancer screenings. First, the researchers split the participants into two groups based on risk. Those in the high-risk group had double readings by radiologists. However, only one radiologist reviewed the results of low-risk patients. That person used an AI decision-support tool that highlighted suspicious characteristics.
This approach caused a 44% decrease in radiologist workload and detected 20% more cancer cases. When the researchers assessed the typical time required to examine mammogram results, they found the AI-assisted option allowed technicians to read approximately 40,000 screenings in the equivalent of five fewer months than usual.
However, the team cautioned this study occurred in Sweden at a single facility. They expressed the need to see if they get similar results in other settings. Their next step is to calculate how AI does at finding cancers during planned screenings versus between mammograms.
The ultimate goal is to detect clinically significant cases earlier without scheduling screenings too frequently. Achieving this would represent a meaningful step in prognoses. Research conducted elsewhere about the nine most common cancers indicated 80% of patients were still alive at least a decade after their diagnoses if health professionals found the disease during stage one or two.
Assessing Breast Density
One reason periodic mammograms for non-symptomatic patients typically begin around age 40 is breast density generally decreases with age, with the most considerable changes occurring during menopause. High-density breast tissue increases cancer risks by interfering with mammogram sensitivity.
Breast cancer density assessments usually occur on a four-point scale through visual assessments of two-view mammograms. However, those checks have numerous limitations. A study suggests AI was 89% accurate at categorizing breast density.
Most U.S. states have laws requiring patients to receive breast density notifications after tests. Those alerts clarify that dense breast tissue is not abnormal, but it makes some cancer detection methods less effective. People can then use that information during future conversations with their providers about whether to receive additional screenings.
The researchers trained the AI breast density tool under radiologists’ supervision. Those technicians then externally validated it. This approach shows how human oversight will remain critical to successful artificial intelligence applications. The technology can increase diagnostic accuracy, but only if it uses reliable data.
Increasing Provider Confidence
A significant limitation of many AI applications in health care and other industries is users cannot see how the tools made specific conclusions. That shortcoming is not always a major concern when artificial intelligence assists with non-critical decisions, but cancer diagnosis outcomes could change a patient’s life. Technological errors waste facility resources, too.
One research team built the first AI tool for breast cancer screenings that shows how it arrived at the results. The participants believed medical professionals would put more trust in tools that provided additional transparency.
They trained their algorithms on 1,136 images from 484 patients within a single health system. The first step was to teach the AI to recognize suspicious lesions while ignoring all irrelevant data. Next, radiologists labeled the images and trained the AI to focus on the lesions’ edges. When the edges have radiating lines or fuzziness, those are among the most reliable indicators of cancerous tissue.
Once the researchers began testing their AI tool, they found it did not surpass radiologist’s assessments but performed as well as the artificial intelligence tools that cannot explain results. Even the most advanced tools can make mistakes, but the critical aspect of this one is people can determine the root causes of errors.
In their future work, the researchers hope to train AI to detect other characteristics of potentially abnormal breast tissue. They would also like to study if the tool assists those analyzing radiologic readings and if it increases their confidence.
A Brighter Future for Breast Cancer Diagnoses
Breast cancer fatalities have decreased over the years, but this disease is still a leading cause of death among females. On the positive side, it has a high survival rate, especially if caught early.
Mammograms and other screening tools are essential for supporting health care professionals in finding cancer causes and determining the best actions for individual patients. AI-powered innovations — including those mentioned here — will not replace human oversight, but many can supplement professional expertise, furthering medical progress.