Internal carotid artery stenosis and occlusion based on machine learning: A comparative analysis of high-resolution MRI and CT angiography

Objective: To explore the clinical application value of the imaging model of carotid artery stenosis occlusion based on machine learning. Methods: A comprehensive analysis was conducted on the medical records, MRI, and CT angiography images of 188 patients diagnosed with severe carotid artery stenosis. Specifically, the training cohort comprised 131 patients, including 107 males and 24 females, with an average age of 68 ± 8 years. A distinct verification group included 57 patients, 50 males, and 7 females, averaging 67 ± 8 years old. For the study, the volume of interest was manually outlined slice by slice, following the contour of the carotid plaque on cross-sectional images. The Radiomics package of Python software was used to extract image omics features. Feature screening was performed using intra - and inter-group correlation coefficients, redundancy analysis, minimum absolute contraction, and selection operator (LASSO) regression analysis. Results: Four image omics features were selected based on the training set to construct the prediction model. Among the six machine learning models, the logistic regression model showed high and stable diagnostic efficiency, with an AUC of 0.872, sensitivity of 100.0%, and specificity of 66.2% in the training set. The AUC of the validation set was 0.867, the sensitivity was 83.3%, and the specificity was 78.8%. The calibration curve and DCA show that the logistic regression model has a reasonable degree of calibration and clinical application value. Conclusion: Predictive MRI and CT angiography models based on machine learning have their respective application values in risk stratification of patients with severe asymptomatic carotid stenosis