Houston Methodist team uses artificial intelligence to predict breast cancer risk

Houston Methodist researchers are using artificial intelligence to quickly and accurately interpret patient charts to help physicians predict breast cancer risk. This could help cut down on unnecessary testing procedures performed as a result of false-positive mammograms.

The computer software evaluates mammograms and pathology reports at 30 times human speed and with 99% accuracy, according to a statement. In a study published Monday in Cancer, Stephen Wong and Dr. Jenny Chang used it to assess the breast cancer risk of 500 breast cancer patients. The software collects diagnostic features, scans patient charts and correlates mammogram results with breast cancer subtypes, according to the statement.

"This software intelligently reviews millions of records in a short amount of time, enabling us to determine breast cancer risk more efficiently using a patient's mammogram. This has the potential to decrease unnecessary biopsies," says Wong, chair of the Department of Systems Medicine and Bioengineering at Houston Methodist Research Institute, in the statement. During the study, the manual review of 50 patient charts by two doctors took 50 to 70 hours. The software churned through all 500 charts in a few hours.

According to the American Cancer Society, about 50% of women who have annual mammograms over a 10-year period will have a false-positive finding. Patients are generally recommended for further testing in the form of an invasive breast biopsy. The American Cancer Society estimates that, of the 1.6 million breast biopsies performed in the U.S. each year, about 20% are unnecessary.

Meanwhile, Harvard and MIT scientists have developed a system to analyze breast cancer pathology that uses artificial intelligence. While the software was 92% accurate and human pathologists were 96% accurate, combining the two methods yielded 99.5% accuracy in detecting cancer. Samsung is also applying deep learning to breast cancer analysis, but is working on computer-aided solutions for ultrasound. Ultrasound images are evaluated in real time, are lower resolution and have more noise, and so are more challenging to accurately process.

- here's the release