Texture Analysis Based on Vascular Ultrasound to Identify the Vulnerable Carotid Plaques

Abstract

Vulnerable carotid plaques are closely related to the occurrence of ischemic stroke. Therefore, accurate and rapid identification of the nature of carotid plaques is essential. This study aimed to determine whether texture analysis based on a vascular ultrasound can be applied to identify vulnerable plaques. Data from a total of 150 patients diagnosed with atherosclerotic plaque (AP) by carotid ultrasound (CDU) and high-resolution magnetic resonance imaging (HRMRI) were collected. HRMRI is the in vivo reference to assess the nature of AP. MaZda software was used to delineate the region of interest and extract 303 texture features from ultrasonic images of plaques. Following regression analysis using the least absolute shrinkage and selection operator (LASSO) algorithm, the overall cohort was randomized 7:3 into the training (n = 105) and testing (n = 45) sets. In the training set, the conventional ultrasound model, the texture feature model, and the conventional ultrasound-texture feature combined model were constructed. The testing set was used to validate the model’s effectiveness by calculating the area under the curve (AUC), accuracy, sensitivity, and specificity. Based on the combined model, a nomogram risk prediction model was established, and the consistency index (C-index) and the calibration curve were obtained. In the training and testing sets, the AUC of the prediction performance of the conventional ultrasonic-texture feature combined model was higher than that of the conventional ultrasonic model and the texture feature model. In the training set, the AUC of the combined model was 0.88, while in the testing set, AUC was 0.87. In addition, the C-index results were also favorable (0.89 in the training set and 0.84 in the testing set). Furthermore, the calibration curve was close to the ideal curve, indicating the accuracy of the nomogram. This study proves the performance of vascular ultrasound-based texture analysis in identifying the vulnerable carotid plaques. Texture feature extraction combined with CDU sonogram features can accurately predict the vulnerability of AP.