Nomogram of Intracranial Artery Calcification with Integration of CT-based Radiomics to Identify Culprit Lesions

Background : To develop a machine learning model of intracranial artery calcification (IAC) based on computed tomography (CT) images and assess its value for improved identifying culprit lesions responsible for acute/subacute ischemic cerebral infarction (ASCI). Methods: Patients with intracranial atherosclerotic diseases in the vertebrobasilar artery or intracranial internal carotid artery who underwent vessel wall MRI and head CT examinations at two hospital centers were retrospectively assessed. Each calcified plaque was classified by the likelihood of having caused an ASCI as culprit or non-culprit. Machine learning technique was utilized to automatically select twenty top-ranked features from IAC segmentation and build a model using the logistic regression algorithm with fivefold stratified cross-validation. The added values of radiomic-based score (Radscore) to stenosis and clinical risk factors for identification of culprit lesions were evaluated using area under the curve (AUC). A nomogram was constructed by integrating the Radscore with clinical and imaging covariates. Results: One hundred and thirty-three ASCI patients with culprit plaques were identified in the training set (totally 282 patients), and 36 were identified in the external test set (totally 71 patients). Diabetes, smoking, coronary heart disease, and stenosis were found to be associated with the culprit lesions in the multivariate analysis. The diagnostic performance of Radscore was 0.674 and 0.609 for the training and external test data set. The nomogram, which includes clinical factors, stenosis, spotty calcification, and Radscore, demonstrated moderate values for the discrimination of symptomatic intracranial atherosclerotic lesions, with an AUC of 0.749 in the training set and an AUC of 0.736 in the external test set. Conclusion: Radiomics of intracranial artery calcification in the culprit lesion may provide added value for identifying ASCI beyond stenosis and clinical factors. The nomogram incorporating both conventional and radiomics variables may serve as a potential diagnostic tool for stroke risk assessment in clinical settings. Trial registration: Retrospectively registered.