Monday, December 2 | 10:20 a.m.-10:30 a.m. | M3-SSNR04-6 | S406B
A deep learning-based radiomics model based on dual-energy computed tomography (DECT) may effectively predict symptomatic carotid plaque in asymptomatic conditions, according to findings to be presented in this neuroradiology scientific session.
Jianson Ji, MD, of Lishui Hospital of Zhejiang University in Hangzhou, China, and colleagues analyzed data from 340 patients with carotid atherosclerosis plaques who underwent DECT and MRI examinations. Their study involved building deep-learning radiomics models by reconstructing virtual monoenergetic images (VMI) at 40, 70, and 100 kiloelectron volt (keV).
A comparison of these models developed using VMI 40 keV, 70 keV, 100 keV, 40+70 keV, 70+100 keV, 40+100 keV, and 40+70+100 keV images found that the model based on 40+70 keV DECT yielded the best performance in predicting symptomatic carotid plaques with the area under the curve (AUC) of 0.87 in the test set of 102 patients.
After integrating key clinical and radiological (plaque ulceration, degree of stenosis, and hyperlipidemia), the AUC was improved to 0.94, according to the research. Patients at high risk of symptomatic carotid plaques portended significantly worse overall survival than those at low risk (p = 0.008).
Ji and colleagues noted that the best model was integrated with significant clinical and radiological features selected by univariate and multivariate logistic regression analyses to achieve the most accurate prediction. Bring your questions to learn more.