July 1, 2020
CRISP Consortium Study Evaluates Elucid Bio’s vascuCAP AI Software to Predict Stroke
July 1, 2020—Elucid Bio, manufacturer of the vascuCAP software, announced that its artificial intelligence (AI) technology demonstrated > 70% improvement in the accuracy of stroke prediction compared with a stenosis-based approach to risk-stratification during a 6-year period in the CRISP Consortium study of carotid risk prediction.
The interim results from the CRISP study were presented virtually during Scientific Session 9 at the Society for Vascular Surgery’s conference, SVS Online Summer 2020. The study is led by Brajesh K. Lal, MD, Director of the University of Maryland’s Center for Vascular Research in Baltimore, Maryland. He is also Director of the NIH Vascular Imaging Core Facility at the University of Maryland.
The company stated that the CRISP study seeks to improve prediction of major adverse neurologic events (MANE; stroke, transient ischemic attack, and amaurosis fugax) by applying AI to traditional carotid CTA imaging using the Elucid Bio vascuCAP software. The vascuCAP software has received FDA clearance and CE Mark approval and is available for commercial use in the United States and Europe.
As summarized in the abstract of the presentation, Dr. Lal reported that colleagues from four centers contributed CTA imaging on patients with asymptomatic carotid stenosis. Dr. Lal’s NIH Vascular Imaging Core at the University of Maryland processed and analyzed the images using vascuCAP for histologically validated plaque geometry and tissue composition from the CTA images. Using the imaging data, the investigators tested several high-performance multivariate models for the response variable of MANE.
Testing differing sets of predictors, the investigators identified the model that optimized the area under the receiver operating characteristic curve (area under the curve [AUC]) for the probability of MANE within 6 years. The best predictor (a combination of morphologic and patient features) was compared with the current standard of care (percent diameter stenosis) for MANE. Additionally, the ranked importance of individual morphologic features in the final model was determined.
The combination of morphologic and patient features created a model with an AUC of 0.86 predication probability. This model significantly outperformed the prediction probability of the commonly used clinical parameter of percent stenosis (using the North American Symptomatic Carotid Endarterectomy Trial criteria: stenosis only, AUC of .45). This resulted in a > 70% improvement in the investigators’ ability to predict MANE in patients with asymptomatic carotid stenosis, reported Dr. Lal at the SVS conference.
“These study results demonstrate that the vascuCAP’s AI algorithms for risk stratification in carotid atherosclerosis are a better predictor of stroke than stenosis assessment alone,” stated Dr. Lal in Elucid Bio’s press release. “Implementing this predictive model on asymptomatic patients in the clinical setting could help identify and inform treatment for those at high-risk for future major adverse neurologic events.”