The prevalence of vertebrobasilar artery (VBA) calcification is high in the intracranial artery, with rates of approximately 20% in the elderly and up to 50% in stroke patients (1). The presence of intracranial artery calcification is suggestive of coexisting atherosclerosis and associated with an elevated risk of stroke (2). Increasing studies have suggested location-dependent differences in the development of atherosclerosis in the posterior circulation plaques compared to the anterior circulation plaques (3). An understanding of the characteristics of VBA calcification may aid in improving stroke risk stratification in the posterior circulation.

Calcification, as a component of advanced atherosclerotic plaque, has been suggested to be associated with ischemic vascular events in various arterial beds. Scattered and multiple calcifications were proposed to be correlated with a higher wall shear stress, thus potentially triggering embolic stroke and acute cardiovascular events (4). Intimal-predominant calcification was associated with an increased risk of stroke and plaque instability in intracranial arteries (5, 6). Nevertheless, the interpretation and analysis of calcification morphology remain restricted to the expertise of radiologists. It is essential to use advanced statistical descriptors to facilitate the non-invasive quantification of radiologic data beyond human-eye capabilities.

Radiomics involves texture analysis by extracting thousands of features to model the spatial distribution of voxel gray-level intensity (7). By applying high-order statistics to quantify image heterogeneity, texture analysis has been greatly developed in oncology (8). Recently, radiomics of coronary artery calcium outperformed the conventional Agatston score in predicting major adverse cardiovascular events on cardiac computed tomography (CT) (9). Therefore, we hypothesized that radiomics of intracranial artery calcification contain more information beyond what can be visually observed or manually quantified.

High-resolution vessel wall magnetic resonance imaging (VW-MRI) allows for the identification of intracranial atherosclerotic plaque that is responsible for recent ischemic events (10). Interpreting calcification can be challenging on routine MRI due to insufficient resolution to detect hypointensity, which is usually surrounded by soft plaque components with complex signal characteristics. CT remains the preferred modality for better visualization and characterization of calcifications. In acute stroke patients, CT is commonly used as an initial diagnostic tool for a shorter acquisition time and greater availability. Combining CT with VW-MRI images enables a comprehensive assessment of plaque calcification, which is easy to implement in the vertebrobasilar artery due to its linear trajectory above the foramen magnum. The purpose of this study was to assess the performance of CT-based texture analysis, as compared to conventional visual assessments, in identifying culprit VBA plaques using VW-MRI as a reference standard.

In this study, we performed a three-dimensional texture analysis to determine whether CT-based radiomic features could be used to discriminate culprit lesions. We also intended to evaluate the diagnostic performance of texture analysis from VBAC compared with conventional methods. Our study found multiple calcification texture parameters differed significantly between culprit and non-culprit lesions. A model with a subset of 5 features according to the minimal-redundancy-maximal-relevance criterion achieved an AUC of 0.81 and an accuracy of 0.73, which significantly outperformed VBAC visual assessments.

To our knowledge, this was the first study to assess the association between radiomic features of intracranial artery calcification and plaque instability using multi-modality imaging methods. Several MRI-defined vulnerable plaque characteristics, such as intraplaque hemorrhage (17) and lipid-rich necrotic core (18), have been validated by specimens from carotid endarterectomy. Nevertheless, current challenges in MR imaging intensity features of intracranial plaque components included spatial resolution and histopathologic validation (19). Vessel wall calcification can be objectively detected on CT images owing to its high attenuation (20). CT texture analysis of vertebrobasilar artery calcification may serve as a promising imaging marker of plaque instability, thereby assisting in clinical decision-making and improving patient outcomes in the posterior circulation stroke.

Several studies have investigated the visual assessment of plaque calcification in relation to stroke risk. Multiple calcifications were associated with the presence of carotid intraplaque hemorrhage, therefore implicating an increased plaque vulnerability (21, 22). A higher prevalence of spotty calcification in the cervicocerebral artery was reported in stroke patients compared with controls (14). A study has found that spotty calcification in nonstenotic carotid atherosclerosis is associated with ischemic stroke on the same side (4). Other studies have demonstrated the pattern of calcification within the intimal layer was associated with unstable plaque phenotype (6) and increased stroke risk (5). In the current study, multiple calcifications had a low sensitivity of 28%, and spotty calcification and intimal predominant calcification have relatively low specificity (54 and 46%) for the presence of culprit lesions. These performances may be explained by the simultaneous development of atherosclerotic calcification in the intimal layer and non-atherosclerotic calcification in the internal elastic lamina and medial layers. Hence, visual assessments of calcification based on its morphology may not provide sufficient data to determine plaque instability.

Radiomics uses computational methods to extract thousands of quantitative features for comprehensive calcification description, mostly imperceptible to the human eye (23). In the animal model of hyperlipidemic mice, several radiomic features of vascular calcium were associated with aging and Western diet, which cannot be determined from calcium scores from conventional CT or total calcium content (24). In the clinical setting, a radiomic-based model was developed from coronary artery calcium radiomic features, resulting in significant improvement in predicting individuals at risk for clinical events (9). In line with previous research, our study demonstrates the significance of calcification radiomics features that extend beyond conventional visual evaluations, achieving the highest AUC values. The diagnostic performance of texture analysis in identifying culprit lesions exhibited a specificity of 90%. This high level of true negative rate suggests that the radiomic model may be better at identifying non-culprit lesions than culprit ones.

Of the 5 selected features, LLH_GLCM_InverseDifferenceMoment in wavelet-filtered image, GLDM_DependenceVariance in 3 mm-sigma LoG-filtered image, and firstorder_ 90Percentile in 4 mm-sigma LoG-filtered image have the significant effect on the radiomic model. The lower value of firstorder_90Percentile in the culprit lesion implies a possible association between low calcification density and plaque instability. GLCM_InverseDifferenceMoment is a measure used to evaluate the homogeneity of an image. GLDM_DependenceVariance is a metric that measures the degree of asymmetry in the distribution of values around the mean value. The above two features suggest that calcification within culprit plaques is more uniform and homogeneous. These variations in texture analysis may offer insights into the underlying pathophysiology of calcium deposits. The lesion, consisting primarily of a specific calcium salt, may demonstrate greater homogeneity compared to the mixed type and is associated with the unstable plaque phenotype (25).

Consistent with previous studies, calcification presence in the proximal intracranial artery (e.g., basilar artery) is relatively rare compared with the distal one (e.g., vertebral artery) (26). Moreover, a higher portion of basilar artery calcification was present in the culprit lesions than in the non-culprit ones. A significant discrepancy was observed in the pattern of calcification within the intracranial arteries (27). A previous study found that calcification in the middle cerebral artery is less pronounced on the symptomatic side compared to the asymptomatic side on CT images (26). In the intracranial internal carotid artery, no difference was detected in the presence of calcification between the symptomatic and asymptomatic sides (28). In a study involving seven intracranial arteries, the presence of calcification is associated with mortality and vascular events in ischemic stroke patients (29). A larger sample size would be necessary to further investigate the relationship between basilar artery calcification and plaque instability.

This study has several limitations. First, the study is retrospective, and the possibility of selection bias cannot be avoided. Second, the sample size of enrolled patients was limited. Texture analysis exhibits significant sensitivity to technical factors, including CT acquisition parameters (30). Therefore, all CT scans in our study were conducted using the same scanning to prevent bias in the extraction of radiomics features. Third, the ROI of calcification is small. We utilized a three-dimensional texture analysis to assess calcification texture, which provided superior performance compared to a two-dimensional approach (31). Fourth, the mRMR algorithm may be overfitted due to the absence of a distinct validation cohort. Despite this, we employed a tenfold cross-validation technique to assess the classifier’s performance (16), improving the generalizability of the model.

CT texture analysis provides an objective evaluation of calcification heterogeneity and outperforms visual assessments in identifying culprit plaques, thereby serving as a potential predictor for plaque instability in the vertebrobasilar artery.

The original contributions presented in the study are included in the article/Supplementary material, further inquiries can be directed to the corresponding author.

The studies involving humans were approved by Shandong Provincial Hospital Affiliated to Shandong First Medical University. The studies were conducted in accordance with the local legislation and institutional requirements. Written informed consent from the patients/participants or patients/participants’ legal guardian/next of kin was not required to participate in this study in accordance with the national legislation and the institutional requirements.

BL: Investigation, Methodology, Writing – original draft. CX: Formal analysis, Software, Writing – original draft. HL: Resources, Visualization, Writing – original draft. CW: Data curation, Project administration, Writing – original draft, Writing – review & editing. SD: Formal analysis, Methodology, Software, Writing – original draft. HY: Conceptualization, Funding acquisition, Resources, Supervision, Writing – review & editing.