Introduction: This work investigates the hypothesis that combined biomechanical and radiomic features (mechano-radiomics) from magnetic resonance (MR) images can help evaluate carotid plaque vulnerability better than conventional MR methods. Hemodynamic forces have long been associated with the destabilization of developed plaques near regions of complex blood flow e.g., the carotid bifurcation, which results in high wall shear stress. High structural stresses within the artery wall caused by the blood flow are directly associated with risk of plaque rupture. Radiomics is an emerging technique to evaluate tissue heterogeneity. Whilst extensively investigated in oncology, there is limited application of these techniques in MR carotid plaque characterisation. Radiomic features may provide better phenotyping of carotid plaques. Mechanics and radiomic analysis were compared against stenosis measurements and morphological plaque characteristics such as the volume of lipid core, intra-plaque haemorrhage (IPH) and calcification.

Methods: Two-dimensional, multi-contrast carotid MR images of 110 patients from the CARE-II multi-centre carotid imaging trial (ClinicalTrials.gov Identifier: NCT02017756) were used to perform the analysis – 55 for the radiomics analysis, and 87 for the 3D modelling of the carotids. Standard radiomic analysis was used to compute over 100 radiomic features derived from each of the multi-slice 2D image stacks. The 2D images were pre-segmented into five plaque components and incorporated into the 3D models. Finite element analysis (FEA) was performed to quantify the wall structural and shear stresses under patient-specific physiological conditions. Radiomic feature analysis was related to patient outcome (culprit or non-culprit lesion) using logistic regression, and the radiomic predictive power was combined with MR-identified plaque components and stenosis measurements to assess the improvement over conventional MR methods. The vessel wall structural and shear stresses were compared between culprit and non-culprit carotid arteries. The robustness of the long-held opinion that culprit carotid arteries (cause of plaque rupture and subsequent TIA or stroke) are associated with higher wall stresses was investigated over a range of material properties.

Results: Combining radiomics biomarkers with a conventional model comprised of geometric and plaque compositional metrics had higher predictive performance over the conventional model alone (area under the curve (AUC) of 0.819±0.002 and 0.689±0.019 respectively, p=0.014). Features within the Grey Level Co-occurrence Matrix (GLCM), Grey Level Dependence Matrix (GLDM) and Grey Level Size Zone Matrix (GLSZM) sub-types were found to be particularly useful in identifying textures which could identify vulnerable carotid plaques. In the mechanics analysis, the patient cohort was stratified against the degree of stenosis. In the group with mild to moderate stenosis ( 70 %), the maximum stress-P1 was consistently higher at critical locations relating to plaque rupture. Locations at the fibrous cap, plaque shoulder and the bifurcation region were considered to evaluate the critical stress. The hypothesis of higher stress on the culprit side was proved to be robust to different material parameters. Evaluations of wall shear stress and wall curvature in three dimensions were also predictors of the culprit side.

Discussion and Conclusions: The results of show that MR radiomics improves upon conventional MR methods for distinguishing culprit and non-culprit plaques. Mechanics adds value by providing a dynamic analysis which combines the 3D artery geometry with patient-specific blood pressures. Recommendations are made for which multi-slice radiomic features should be used in terms of robustness to variations in carotid segmentations and their predictive ability. Radiomic analysis helps identify texture features of vulnerable plaque phenotypes, such as lipid-rich necrotic core or calcification distribution, while biomechanical analysis proved that over different material properties, the stress was found to be higher on culprit sides at critical locations which included the carotid bifurcation and multiple atherosclerotic plaques. A mechano-radiomic model for assessing the vulnerability of carotid plaque is then proposed to obtain the maximum benefit from both a mechanics and radiomic analysis.}