Artificial Intelligence Software Speeds Up Breast Cancer Risk Prediction, Resolves False-Positives

Artificial Intelligence Software Speeds Up Breast Cancer Risk Prediction, Resolves False-Positives

Artificial intelligence (AI) software developed by a team of researchers at Houston Methodist Hospital is designed to reliably interpret mammogram data and enable doctors to quickly and accurately assess the likelihood of breast cancer risk.

The Houston Methodist investigators reported their progress in a new study published online before print in the journal Cancer, explaining that the computer software intuitively translates patient charts into diagnostic information 30 times faster than would be required by a typical human analyst, and with 99 percent accuracy.
The original article, Correlating mammographic and pathologic findings in clinical decision support using natural language processing and data mining methods, observes that a key challenge confronting researchers mining electronic health records such as mammography data is a preponderance of unstructured narrative text which significantly limits usable output.

The authors note that breast cancer subtype imaging characteristics have been described previously, but without any standardization of parameters for data mining.


In their investigation, the authors searched Houston Methodist Hospital’s enterprise-wide data warehouse — the Methodist Environment for Translational Enhancement and Outcomes Research (METEOR) — for case histories of patients with Breast Imaging Reporting and Data System, or BI-RADS, (a standard method of reporting mammogram results) category 5 mammogram readings between January 2006 and May 2015 and available pathology reports.

They developed natural language processing (NLP) software algorithms that can automatically extract findings from free text mammogram and pathology reports. Then they analyzed correlations between mammography imaging features and breast cancer subtypes using one-way analysis of variance and the Fisher exact test — a statistical significance test used in analysis of contingency tables to determine if there are nonrandom associations between two categorical variables, and typically employed when sample sizes are small.

WongSThe investigative team, led by Stephen T. Wong, PhD, PE, chairman of the Department of Systems Medicine and Bioengineering at Houston Methodist Research Institute, and Dr. Jenny C. Chang, MD, director of the Houston Methodist Cancer Center, used the AI software to scan patient charts, collect diagnostic features, and evaluate mammograms and pathology reports of 543 breast cancer patients, determining that their NLP algorithm was able to obtain key characteristics for patients who met the inclusion criteria.

They conclude that mammography imaging characteristics obtained using NLP techniques for conducting automated text ChangJsearch and extraction of mammogram reports that correlate with pathologic breast cancer subtypes validate previously reported trends assessed by manual data collection.

Clinicians were able to use results like the expression of tumor proteins to accurately predict each patient’s probability of breast cancer.

“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,” Wong said in a Houston Methodist press release.

While a manual review of 50 charts would typically take two clinicians roughly 50 to 70 hours to complete, the Houston Methodist team’s AI software was able to review 500 charts in a few hours, saving more than 500 physician hours. “Accurate review of this many charts would be practically impossible without AI,” Wong said.

According to the Centers for Disease Control and Prevention (CDC), 12.1 million mammograms are performed
annually in the U.S. A whopping 50 percent of those scans yielding false positive results, according to the American Cancer Society (ACS). This means one in every two healthy women who get mammograms are being unnecessarily subjected to the trauma of being told they likely have cancer.

The Houston Methodist researchers note that currently, when mammograms fall into the “suspicious” category — a broad range of 3 to 95 percent cancer risk — patients are advised to get a breast biopsy. More than 1.6 million of them are performed annually nationwide, and about 20 percent of them are unnecessary due to false-positive mammogram results, according to ACS estimates.

The Houston Methodist team hopes their artificial intelligence software will serve as a tool to help physicians more precisely define what percentage of risk makes a biopsy appropriate, thereby decreasing unnecessary breast biopsies.

The research was supported in part by the Houston-based John S. Dunn Research Foundation, a private foundation dedicated to supporting health programs, medical research, and healthcare education organizations in the greater Houston area.