Integrating MRI radiomics with novel fluid-based texture features (GLFZM) to predict atherothrombotic stroke risk.

Purpose: The purpose of this study was to evaluate the predictive value of MRI-based texture features for assessing stroke risk from vulnerable carotid plaques.

Method: Among patients diagnosed with carotid artery plaque by MRI, 10 patients with whom Time-to-Event for atherothrombotic stroke could be obtained were enrolled. Radiomics features were extracted from T1/T2-weighted black-blood images and cervical 3D time-of-flight images. Additionally, this investigation employed the extraction of 16 Gray-Level Fluid Zone Matrix (GLFZM) features, specifically developed for this analysis. Wall shear stress (WSS), a biomechanical characteristic, was also subjected to calculation. These features served as the basis for developing clinical models, radiomics-plaque models, radiomics-lumen models, GLFZM models, WSS models, and combined models. The performance of each model was evaluated using regression analysis by calculating mean squared error (MSE). As one aspect of the robustness of each model, we evaluated the models using Cox proportional hazard models and concordance indices (CI) derived from synthetic data generated with the noise scale.

Result: The LOOCV MSE and mean CI values were: clinical model (2.58 × 10⁶, 0.65), radiomics-plaque model (4.62 × 10⁶, 0.75), radiomics-lumen model (3.30 × 10⁶, 0.81), GLFZM model (2.00 × 10⁶, 0.74), WSS model (2.47 × 10⁶, 0.46), and combined model (1.48 × 10⁶, 0.78). The combined model demonstrated the minimal MSE.

Conclusion: This study demonstrated via preliminary simulations that analyzed clinical variables, radiomic features (plaque and lumen), texture features indicative of flow velocity (GLFZM), and biomechanical features (WSS) as model predictors, the potential utility of texture analysis in forecasting ischemic events in cerebral infarction resulting from vulnerable carotid plaques.