Establishment and validation of a risk stratification model for stroke risk within three years in patients with cerebral small vessel disease using a combined MRI and machine learning algorithm

Abstract

BackgroundCerebral small vessel disease (CSVD) is a major cause of stroke, particularly in the elderly population, leading to significant morbidity and mortality. Accurate identification of high-risk patients and timing of stroke occurrence plays a crucial role in patient prevention and treatment. The study aimed to establish and validate a risk stratification model for stroke within three years in patients with CSVD using a combined MRI and machine learning algorithm approach. MethodsThe assessment encompassed demographic, clinical, biochemical, and MRI-derived parameters. Correlation analysis, logistic regression, receiver operating characteristic (ROC) curve analysis, and nnet neural network algorithm were employed to evaluate the predictive value of machine learning algorithms and MRI parameters for stroke occurrence within 3 years in patients with CSVD. ResultsMRI-derived parameters, including average WMH volume, perfusion deficit volume, ischemic core volume, microbleed count, and perivascular spaces, exhibited strong correlations with stroke occurrence (P < 0.001). MRI-derived parameters demonstrated high sensitivities (0.719 to 0.906), specificities (0.704 to 0.877), and AUC values (0.815 to 0.871). The combined model of machine learning algorithms and MRI parameters yielded an AUC value of 0.925, indicating significantly high predictive accuracy for identifying the risk of stroke within three years in CSVD patients. ConclusionThe integrated risk stratification model, incorporating machine learning algorithms and MRI parameters, demonstrated strong predictive potential for stroke within three years in patients with CSVD. This model offered valuable insights for personalized interventions and clinical decision-making in the management of CSVD.