High-resolution MRI-based Radiomics and Machine Learning for Identification of Middle Cerebral Artery Vulnerable Plaques.

INTRODUCTION

To construct a vulnerable plaque model utilizing high-resolution magnetic resonance imaging (HR-MRI) based radiomics features and traditional features.

MATERIALS AND

METHODS

110 patients with unilateral middle cerebral artery (MCA) stenosis and available HR-MRI data were enrolled. 50 plaques exhibiting acute stroke signals within the stenotic MCA region were categorized into the vulnerable plaque group, whereas the remaining 60 plaques were assigned to the stable plaque group. A traditional model was constructed by calculating the radiological plaque features. Radiomics features from HR-MRI were employed to establish a vulnerable plaque model with a support vector machine (SVM). A combined predictive model was then built. The efficacy of these models was assessed using receiver operating characteristic curves.

RESULTS

The traditional model demonstrated an AUC of 0.852 in the training set and 0.839 in the test set. While the radiomics model and the combined model showed an improved AUC: 0.919 vs. 0.902 for the training sets and 0.955 vs. 0.949 in the test sets. Both the radiomics model and the combined model outperformed the traditional model in the two sets (P<0.05); the performance of the combined model was not significantly different from that of the radiomics model in both sets (P>0.05).

DISCUSSION: A combined model based on radiomics and traditional features improves the ability to differentiate the vulnerable MCA plaques from stable ones.

CONCLUSIONS

The radiomics model derived from HR-MRI exhibits high accuracy in identifying the vulnerable plaques that are likely to cause acute stroke, and this model was significantly superior to the traditional model.