The Journal of the American College of Radiology published an encouraging artificial intelligence (AI)-related study in mid-August. The study, which was conducted at Harvard Medical School and led by Vladimir Valtchinov, PhD, showed that AI could be used to evaluate digital clinical reports and accurately pick out patients who wore implantable devices that could be a safety issue if they underwent an MRI.
The study compared two distinct approaches of techniques known as natural language processing (NLP). The first was an ontology-derived approach that pulls out medical terms based on public biomedical ontologies, and the second was an “expert”-originated method that was preloaded with applicable search terms.
The experimental results showed that both AI approaches resulted in high accuracy when attempting to determine whether or not patients with implantable devices were of high safety risk for MRI; the approaches analyzed various clinical notes such as emergency department notes and radiology reports. For the ontology-derived approach, the sensitivity was 82%, specificity 90%, and accuracy 83%. For the expert-derived approach, the sensitivity was 88%, specificity 96%, and accuracy 91%.
“In a clinical workflow when an MR imaging is ordered, a trigger to analyze clinical records before the examination would be helpful to inform providers and patients about potential safety risks,” Valtchinov et al wrote. “This step could be added either before patient arrival or right before an examination.”
The team believes that in the future, they may wish to conduct studies that look into using AI to analyze automated report retrieval methods.