Multidimensional clustering of clinical, metabolic, and ultrasound features for 30-day mortality risk stratification in ischemic stroke patients with carotid atherosclerotic plaque

Figures

• 

  Figure 1. Technical roadmap for risk stratification and subgroup analysis of ischemic stroke patients with carotid plaques via multi-dimensional feature clustering. This flowchart outlines the study workflow: 13,568 patients were screened, with 3,764 eligible participants enrolled after applying inclusion and exclusion criteria. Multi-dimensional data were collected and preprocessed. The optimal cluster number was determined via the elbow method and silhouette score, followed by UMAP dimensionality reduction and k-means clustering to identify five distinct subgroups. Inter-subgroup differences were analyzed using appropriate statistical tests, with visualization of key indicators. Finally, subgroup-specific characteristics were summarized to guide personalized clinical interventions.

• 

  Figure 2. Elbow method and silhouette coefficient curves. This figure displays two clustering optimization curves plotting the number of clusters (k-value) against sum of squared errors (SSE) and silhouette coefficient. The elbow point and maximum silhouette coefficient both appear at k = 5, indicating that five clusters represent the optimal number for patient stratification in this dataset.

• 

  Figure 3. UMAP dimensionality reduction and cluster visualization. This figure presents a two-dimensional visualization of high-dimensional patient data using UMAP algorithm. The five distinct colored clusters (Cluster 0–4) show spatial separation in the reduced dimensional space, demonstrating successful clustering of patients with different clinical profiles into distinct subgroups.

• 

  Figure 4. Demographic and clinical characteristics across clusters. This figure displays bar charts or grouped visualizations comparing key demographic and clinical parameters including age, gender distribution, NIHSS score, and other baseline characteristics across the five patient clusters, allowing visual comparison of phenotypic differences between subgroups.

• 

  Figure 5. Laboratory and inflammatory markers across clusters. This figure presents comprehensive visualization of inflammatory and laboratory markers such as WBC, CRP, and CK across clusters using multiple subplots. The inter-cluster variations in these acute phase reactants and inflammatory biomarkers are clearly displayed.

• 

  Figure 6. Metabolic indicators part 1 across clusters. This figure displays box-and-whisker plots or similar visualizations of metabolic parameters including HbA1c, LDL cholesterol, triglycerides, and HDL cholesterol across five clusters, showing differences in glucose and lipid metabolism among patient subgroups.

• 

  Figure 7. Metabolic indicators part 2 across clusters. This figure presents additional metabolic and hepatorenal function markers including ALT, AST, creatinine, and urea across clusters, visualizing liver and kidney function status and metabolic parameters in different patient subgroups.

• 

  Figure 8. Coagulation indicators across clusters. This figure displays coagulation parameters including INR, PT, aPTT, D-dimer, and fibrinogen across five clusters, illustrating inter-cluster variations in thrombotic tendency and coagulation function.

• 

  Figure 9. Blood gas indicators across clusters. This figure presents arterial blood gas parameters including PaCO2, PaO2, pH, and HCO3− across clusters, reflecting respiratory and metabolic acid–base status differences among patient subgroups.

• 

  Figure 10. Ultrasound plaque characteristics across clusters. This figure comprehensively displays carotid plaque characteristics including maximum thickness, length, stenosis rate, echogenicity pattern, surface regularity, surface integrity, and Plaque-RADS score across five clusters, showing substantial inter-cluster variations in plaque morphology and stability features.