by John R. Fischer
, Senior Reporter | October 22, 2021
Designed to identify patterns across thousands of breast ultrasounds, a new AI tool may help confirm breast cancer diagnoses without the need for biopsies and tissue samples, say researchers in New York City.
Teams at NYU Langone Health and its Laura and Isaac Perlmutter Cancer Center trained their AI software on what they believe is the largest data set of its kind. The set is made up of 288,767 separate ultrasound exams taken from 143,203 women treated at NYU Langone hospitals.
When tested separately on 44,755 completed ultrasound scans, their solution was able to help radiologists identify breast cancer 37% more and reduced the number of tissue samples or biopsies required for confirming suspected cases by 27%.
“Our study addresses one of the limitations to increase utilization of screening ultrasound in addition to screening mammography — the substantial reduction of the false positive findings detected on screening ultrasound,” co-investigator and radiologist Dr. Linda Moy told HCB News.
One in eight women in the U.S. will be diagnosed with breast cancer in their lifetime, with more than 300,000 positive diagnoses made in 2021 alone, according to the American Cancer Society.
Despite its high rate for false positives, screening ultrasound can be used as a supplemental tool in women with dense breast tissue, which often obscures and makes it difficult for mammography to detect tumors, due to its low sensitivity. Ultrasound also does not involve exposure to radiation and is cheaper than other modalities like MR, making it more widely available in community clinics, says Moy.
In the study, over half of ultrasound breast exams were used to create the computer program. Ten radiologists each reviewed a separate set of 663 breast exams, with an average accuracy of 92%. The addition of the AI tool brought this average to 96%, with all diagnoses checked against tissue biopsy results.
Senior investigator Krzysztof Geras says that he and his team plan to refine the AI software with additional patient information, such as a woman’s added risk based on family history or genetic mutation, which was not included in their latest analysis. He also says the analysis only evaluated past exams and that clinical trials in current patients and real-world scenarios are required.
“Our first step is going to be evaluating the feasibility of deploying our model clinically at NYU Langone,” he told HCB News. “If we find a suitable partner for commercialization of our AI, we are going to pursue FDA clearance in the future. We believe that our technology is already advanced enough that it has a potential to have a positive impact on the quality of diagnosis when used to assist radiologists.”
The findings were published in Nature Communications