Author + information
Type 2 diabetes has been recognized as a major risk factor for cardiovascular disease. Several studies have indicated that type 2 diabetes was associated with plaque morphology and composition. The purpose of this study was to investigate whether type 2 diabetes can be identified based on carotid ultrasound plaque texture features extracted by MaZda in atherosclerotic patients.
A total of 92 participants (diabetics 46 and non-diabetics 46) with 192 carotid plaques (diabetics 100, non-diabetics 92) were included. The plaques were classified into echo-rich (n = 65), intermediate (n = 53) and echolucent (n = 74) plaques according to the ctiteria of the European carotid plaques study group. The ability of each type of plaque for identifying type 2 diabetes was examined. In MaZda, six different texture feature sets were extracted from the ultrasound images of carotid plaques using the following algorithms: histogram, absolute gradient, run-length matrix, gray-level co-occurrence matrix, autoregressive model and wavelet transform. Thirty optimal features were selected using Fisher coefficient and mutual information measure, respectively. The logistic regression analysis was performed to classify the diabetics and the non-diabetics based on the optimal feature sets. The diagnostic validity of the model output (predicted probability) was further tested with the receiver operating characteristic (ROC) curves.
Visual classification of plaques showed a good agreement between two observers and the inter-observer reproducibility was 96.96% (kappa value = 0.954). When logistic regression analysis was carried out based on the texture features of echo-rich, intermediate and echolucent plaques, the area under the ROC curve was 0.739, 0.852 and 0.771, respectively (p < 0.001).
Carotid plaque texture analysis may be a non-invasive and promising predictor for type 2 diabetes in atherosclerotic patients.