A computer program trained to see patterns among thousands of breast ultrasound images can help doctors accurately diagnose breast cancer, according to a new study.
Tested separately on 44,755 ultrasound exams already performed, the artificial intelligence (AI) tool improved the ability of radiologists to correctly identify disease by 37% and reduced the number of tissue samples, or biopsies, necessary to confirm suspicious tumors.
Led by researchers from NYU Langone Health’s Department of Radiology and its Laura and Isaac Perlmutter Cancer Center, the team’s AI scan is believed to be the largest of its kind, involving 288,767 separate ultrasound exams performed out of 143,203 women treated at NYU Langone hospitals in New York between 2012 and 2018. The team’s report is published online September 24 in the journal Nature Communication.
Our study shows how artificial intelligence can help radiologists who read breast ultrasounds reveal only those with real signs of breast cancer and avoid biopsy verification in cases that turn out to be mild. “
Krzysztof Geras, PhD, Principal Investigator, NYU Langone Hospitals
Ultrasound exams use high-frequency sound waves passing through tissue to construct real-time images of the breast or other tissue. While not typically used as a breast cancer screening tool, it has served as an alternative (to mammography) or as a follow-up diagnostic test for many women, says Geras, assistant professor in the radiology department. from NYU Grossman School of Medicine and a member of the Perlmutter Cancer Center.
Ultrasound is cheaper, more widely available in community clinics, and does not involve radiation exposure, the researchers say. Additionally, ultrasound is better than mammography at penetrating dense breast tissue and distinguishing compact but healthy cells from compact tumors.
However, the technology has also been found to lead to too many false breast cancer diagnoses, producing anxiety and unnecessary procedures for women. Some studies have shown that a majority of breast ultrasound examinations showing signs of cancer are found to be non-cancerous after biopsy.
âIf our efforts to use machine learning as a triage tool for ultrasound studies prove successful, ultrasound may become a more effective tool in breast cancer screening, particularly as an alternative to mammography, and for people with dense breast tissue, âsays study co-investigator and radiologist Linda Moy, MD. âIts future impact on improving the health of women’s breasts could be profound,â adds Moy, a professor at the NYU Grossman School of Medicine and a member of the Perlmutter Cancer Center.
Geras cautions that while his team’s early results are promising, his team only looked at past reviews in their latest analysis, and clinical trials of the tool on current patients and real conditions are needed before it can be tested. ‘it can be routinely deployed. He also plans to refine the AI ââsoftware to include additional patient information, such as a woman’s additional risk of having a family history or a genetic mutation linked to breast cancer, which was not. included in their latest analysis.
For the study, more than half of the ultrasound examinations of the breasts were used to create the computer program. Ten radiologists then each examined a separate set of 663 breast exams, with an average accuracy of 92 percent. With the help of the AI ââmodel, their average accuracy in diagnosing breast cancer improved to 96%. All diagnoses were verified against tissue biopsy results.
The latest statistics from the American Cancer Society estimate that one in eight women (13%) in the United States will be diagnosed with breast cancer in her lifetime, with more than 300,000 positive diagnoses in 2021 alone.
NYU Langone Health / NYU Grossman School of Medicine