Lipid-related radiomics of low-echo carotid plaques is associated with diabetic stroke and non-diabetic coronary heart disease.

Carotid plaque radiomics-included models have increased the predictive capacity of cardiovascular risk, but the radiomic features of these models were inconsistent in previous studies. Lipids could be used to select the most important radiomic feature. A retrospective case‒control study was performed in 153 diabetic and 76 non-diabetic patients with carotid plaque detected by ultrasound. Cerebro-cardiovascular disease (CCD), comprising coronary heart disease (CHD) and stroke, was the primary outcome. Clinical variables and radiomic features of longitudinal carotid plaque images were collected. Principal component analyses were used to compare the power of radiomic and lipid features in discrimination between diabetes, CCD patients, and their opposites. Partial least square regression, logistic regression analyses, and least absolute shrinkage and selection operator (LASSO) regression were performed for high-risk radiomic features. The diagnostic capacity of the models was evaluated. PCA based on radiomics or lipids did not show good discrimination of diabetes, CCD, and their opposites. There were 6 overlapping radiomic features associated with lipid profiles, but only original_firstorder_Mean was negatively associated with diabetic stroke [adjusted OR = 0.468 (0.243-0.902), P = 0.023] and nondiabetic CHD [adjusted OR = 0.311 (0.123-0.783), P = 0.013]. The associations remained independent in the LASSO regression models (β=-0.032 for diabetic stroke, and - 0.026 for non-diabetic CHD). The diagnostic capacity of lipid-related radiomics for diabetic stroke (0.556 to 0.688) and non-diabetic CHD (0.690 to 0.783) was increased by the combination of these clinical variables. Carotid plaque radiomics is associated with lipids and stroke in diabetes, and quantitative features are useful for therapeutic guidance and cardiovascular risk evaluation in clinical use.