AI technique for mammography shows promise for improved accuracy

By combining readings of clinical reports and X-rays, a team of researchers believes it can significantly reduce unnecessary biopsies.
By Gus Venditto
04:06 PM
Houston Methodist pioneers AI software for mammography

False positives in mammography test results have been a persistent problem in screening for breast cancer. The American Cancer Society has estimated that up to 50 percent of test results may result in a woman requiring additional testing for no reason.

Now, a team at Houston Methodist believes it has developed an artificial intelligence software program that can improve readings to 99 percent accuracy by analyzing values from X-ray images and the text of clinical reports, complementing BI-RADS categories with additional information and parameters. 

The team, led by Stephen T. Wong, chair of the Department of Systems Medicine and Bioengineering at Houston Methodist Research Institute, and Jenny Chang, MD, director of the Houston Methodist Cancer Center, developed a software technique that scans patient charts, collected diagnostic features and correlated mammogram findings with breast cancer subtype. Clinicians used results, such as the expression of tumor proteins, to accurately predict each patient’s probability of breast cancer diagnosis.

"We figured out you can mine a clinical report for additional information," said Wong, "most of the cliniical reports are not in a structured format, they are in free form text. So if we can run an AI program to extract the medical information information and build a risk assessment model we can score the information and reduce unnecessary biopsies."

[Also: Google DeepMind teams up with London hospitals to put machine learning to work against head and neck cancers]

The team had access to over 10,000 mammography and clinical reports in the hospital's data warehouse of BI-RADS 4 defined patient records. The study was funded internally by Houston Methodist.  Their AI algorithm has been validated in BI-RADS 5 (type 5, >95% of breast cancer risks) with over 540 patient records at 99 percent accuracy. The results were recently published in the journal Cancer. 

Wong believes that in the future, once the system is thoroughly vetted, it can become part of the EHR and integrated into clinical workflows, providing clinicians with an assessment of patient risk and recommendations for biopsy type.

According to the Centers for Disease Control and Prevention, more than 12 million mammograms are performed in the U.S. each year. When mammograms fall into the suspicious category, a broad range of 3 to 95 percent cancer risk, patients are recommended for biopsies. That has resulted in more than 1.6 million breast biopsies performed each year. The American Cancer Society estimates that about 20 percent are not necessary.

Twitter: @GusVenditto
Email the writer: gus.venditto@himssmedia.com


Like Healthcare IT News on Facebook and LinkedIn