A new study reveals that a computer program taught to look for patterns among thousands of breast ultrasound images can assist physicians in precisely detecting breast cancer.
When tested independently on 44,755 ultrasound exams that had already been completed, the artificial intelligence (AI) tool enhanced the skill of radiologists to properly identify the disease by 37% and decreased the number of tissue samples, or biopsies, required to validate suspect tumors by 27%.
Directed by scientists from the Department of Radiology at NYU Langone Health and its Laura and Isaac Perlmutter Cancer Center, the AI analysis carried out by the team is said to be the most extensive of its kind, covering 288,767 separate ultrasound exams taken from 143,203 women who were treated at NYU Langone hospitals in New York City from 2012 to 2018.
The research paper has been published online in the September 24th issue of the journal Nature Communications.
Our study demonstrates how artificial intelligence can help radiologists reading breast ultrasound exams to reveal only those that show real signs of breast cancer and to avoid verification by biopsy in cases that turn out to be benign.
Krzysztof J. Geras, PhD, Study Senior Investigator and Assistant Professor of Radiology, Grossman School of Medicine, NYU
Ultrasound exams employ high-frequency sound waves traveling through the tissue to build real-time images of breast or other tissues. Although not commonly used as a breast cancer screening tool, it has been used as an alternative to mammography or follow-up diagnostic tests for a number of women, says Dr. Geras, an assistant professor of radiology at NYU Grossman School of Medicine and a member of Perlmutter Cancer Center.
Ultrasound is economical, more extensively available in community clinics and does not expose the patient to radiation, the scientists state. Furthermore, ultrasound is better than mammography for probing dense breast tissue and differentiating packed but healthy cells from compact tumors.
Yet, the technology has also been proven to result in several false diagnoses of breast cancer, creating anxiety and needless procedures for women. A few studies have revealed that many breast ultrasound exams signifying signs of cancer have proven to be noncancerous after biopsy.
If our efforts to use machine learning as a triaging tool for ultrasound studies prove successful, ultrasound could become a more effective tool in breast cancer screening, especially as an alternative to mammography, and for those with dense breast tissue. Its future impact on improving women’s breast health could be profound.
Linda Moy, MD, Study Co-Investigator and Professor of Radiology, Grossman School of Medicine, NYU
Moy is also a member of Perlmutter Cancer Center.
Dr. Geras warns that while his team’s early results are encouraging, they only studied past exams in their latest investigation, and clinical trials of the tool in current patients and real-world scenarios are required before it can be regularly used.
He is also keen to improve the AI software to include extra patient information, such as a woman’s added risk from genetic mutation tied to breast cancer or having a family history, which was not added in their latest analysis.
For the research, over half of ultrasound breast examinations were used to build the computer program. Ten radiologists then each assessed a separate set of 663 breast exams, with an average precision of 92%. When aided by the AI model, their average accuracy in diagnosing breast cancer improved to 96%. All diagnoses were verified against tissue biopsy results.
The most recent statistics from the American Cancer Society approximate that one in eight women, or 13% of women, in the United States will be diagnosed with breast cancer over their lifetime, with over 300,000 positive diagnoses in just 2021.
The study received funding support from the National Institutes of Health grants P41 EB017183 and R21 CA225175; the Gordon and Betty Moore Foundation grant 9683; the National Science Foundation grant HDR-1922658; and the Polish National Agency for Academic Exchange grant PPN/IWA/2019/1/00114/U/00001.
In addition to Dr. Geras and Dr. Moy, other NYU Langone scientists involved in this research are co-lead investigators Yiqiu “Artie” Shen, Farah Shamout, and Jamie Oliver; and co-investigators Jan Witowski, Jungkyu Park, Kawshik Kannan, Nan Wu, Stacey Wolfson, Alexandra Millet, Connor Huddleston, Robin Ehrenpreis, Cathy Tyma, Naziya Samreen, Divya Awal, Yiming Gao, Stacey Gandhi, Beatriu Reig, Cindy Lee, Chloe Chhor, Sheila Kumari-Subaiya, Cindy Leonard, Christopher Moczulski, Jaime Altabet, Reyhan Mohammed, James Babb, Alana Lewin, and Laura Heacock.
Shen, Y., et al. (2021) Artificial intelligence system reduces false-positive findings in the interpretation of breast ultrasound exams. Nature Communications. doi.org/10.1038/s41467-021-26023-2.