AI-Based Algorithm Significantly Reduce Miss Rates for Pulmonary Embolism on CT Imaging

Pulmonary Embolism (PE) ranks as the third most common acute cardiovascular syndrome globally with a prevalence of 20%. Untreated, PE can have a mortality rate as high as 35%, but this can be reduced to between 2 and 15% with rapid and appropriate treatment. The primary diagnostic tool for PE is CT pulmonary angiography (CTPA), a non-invasive test that is both widely available and quick to perform, capable of detecting emboli with high sensitivity and specificity. However, diagnosing PE remains complex due to the presence of other conditions that can appear similar on scans, such as tumors, lymph nodes, artifacts, and lung nodules. Studies have indicated that delays or gaps in PE diagnosis are a leading cause of preventable deaths due to missed diagnoses. Now, a new study has shown that an artificial intelligence (AI)-based algorithm can substantially reduce the rates of missed PE diagnoses on CT scans.

For the study, researchers from University of California Irvine (Irvine, CA, USA) retrospectively collected 1204 CTPAs from 230 US cities, utilizing 57 different scanner models from six manufacturers. The gold standard, or ground truth, was established by consensus among three US board-certified expert radiologists. These cases were also assessed by an AI algorithm named CINA-PE, designed to detect and identify suspected PE locations. The algorithm’s effectiveness was measured both per case and per-finding.

Analysis included cases where PEs were present but not reported clinically, yet were identified by the AI. Of the 196 confirmed cases, 29 (15.6%) were initially unreported. The AI algorithm successfully identified 22 of these 29 cases, thus reducing the miss rate from 15.6% to 3.8% (7 missed out of 186 cases). These findings, published in the journal Clinical Imaging, suggest that AI integration into clinical settings can enhance the accuracy of PE diagnosis, leading to better patient outcomes through timely treatments. The implementation of such AI tools could significantly reduce the incidence of overlooked or delayed diagnoses, improving overall healthcare delivery and patient care.

Related Links:
University of California Irvine