Trailblazing AI tool tracks cancer via radiology reports
Artificial intelligence (AI) is playing an increasingly important role in healthcare. A recent example of how this type of technology can be applied was displayed in a recent study by researchers at Dana-Farber Cancer Institute. Using 14,000 imaging reports for 1,112 patients, the study aimed to determine whether an AI algorithm could review and analyze the data as efficiently and accurately (or more so) than human reviewers.
The records were first manually reviewed by the study authors and colleagues using the “PRISSMM” framework, the standard in data structuring at Dana-Farber. This allowed the reviewers to analyze the imaging reports and note if cancer was present, whether the cancer was getting worse or improving, and if the cancer had spread to specific sites in the body. A computational model was trained to recognize the same outcomes from the given reports. Then the AI algorithms were applied to an additional 15,000 reports that had not been manually reviewed through human assessment. The researchers found that the computed outcomes predicted survival rates with similar accuracy to the outcomes generated through manual review.
While the human reviewers annotated reports at a rate of about three patients per hour (meaning it would take one human reviewer six months to annotate all 30,000 imaging reports), researchers found that the AI model could annotate 30,000 reports in approximately 10 minutes.
“To create a true learning health system for oncology and to facilitate delivery of precision medicine at scale, methods are needed to accelerate curation of cancer-related outcomes from electronic health records,” said the study authors. Looking ahead, the researchers posited, “this technique could substantially accelerate efforts to use real-world data from all patients with cancer to generate evidence regarding effectiveness of treatment approaches.”