An Integrated Visual Analytics System for Studying Clinical Carotid Artery Plaques

Chaoqing Xu, Zhentao Zheng, Yiting Fu, Baofeng Chang, Legao Chen, Minghui Wu, Mingli Song, Jinsong Jiang

TL;DR

Researchers have developed an integrated visual analytics system that helps vascular surgeons bridge the gap between a patient's clinical history and their medical imaging. By combining machine learning with interactive visualizations, the system allows experts to explore how factors like smoking, age, and complex "radiomic" features influence the formation and composition of dangerous carotid artery plaques.

What Problem Does This Paper Address?

Carotid artery plaques are a major cause of strokes and heart attacks, yet diagnosing them remains a fragmented process. Currently, physicians must mentally synthesize two very different types of data: text-based clinical indicators (like blood pressure, age, and smoking history) and medical images (like CTA or MRA scans).

While medical imaging shows the size and location of a plaque, it often hides subtle "radiomic" features—mathematical patterns in texture and shape—that could indicate how dangerous a plaque actually is. Furthermore, standalone machine learning models designed to classify these plaques often act as "black boxes," lacking the interpretability surgeons need to make informed clinical decisions. This paper introduces a system designed to unify these data streams into a single, interactive exploratory environment.

Methods

The authors designed a Visual Analytics (VA) workflow that integrates data collection, statistical analysis, and machine learning.

  1. Data Extraction: The system pulls clinical recording data (demographics, chronic diseases, and risk factors like troponin levels) and CTA imaging data.
  2. Radiomics Pipeline: Using the "Pyradiomics" plugin for 3D Slicer, the team extracted high-dimensional features from four manually annotated plaque types: Intraplaque Hemorrhage (IPH), IPH with high lipid content, Calcium, and Fibrous tissue.
  3. Machine Learning: A Support Vector Machine (SVM) classifier analyzes these radiomic features to rank their importance and evaluate their correlation with different plaque types.
  4. Interactive Visualization: The front-end consists of four modules:
  5. Data Input: For selecting patient cohorts.
  6. Linked Info Visualization: Uses charts like stacked area plots and scatter plots to show relationships between clinical factors (e.g., BMI, Age) and plaque size.
  7. Multiclass Classification: Displays ML results through parallel coordinates and radar charts to make the "black box" interpretable.
  8. Plaque Rendering: Provides 3D spatial visualizations of the plaques within the artery.
Fig. 2. The UI of our VA system (top) and workflow (bottom), where each analysis step is annotated with the label of the corresponding system module.It includes four main elements: (A) Data Input Module. (B) Linked Information Visualization Module. (C) Multicalss Classification Module. (D) Carotid Artery Plaque Rendering Module.

Datasets

The study utilized a real-world dataset obtained from Zhejiang Provincial People’s Hospital. For the evaluation and case studies, the researchers focused on a cohort of 22 subjects. The data included anonymized clinical records and DICOM image cohorts of patients' head and neck portions following CTA scanning.

Key Insights and Ideas

  • Bridging the Interpretability Gap: By using parallel coordinates to visualize the top 10 features identified by the SVM, the system allows surgeons to see why the model classified a plaque a certain way, rather than just receiving a label.
  • Hidden Correlations: A standout finding was the high correlation between the radiomic feature orig_glrlm_RLNU (Run Length Non-Uniformity) and smoking. This suggests that the impact of smoking on arterial health is physically reflected in specific, measurable texture patterns within the plaque.
  • Plaque-Specific Profiles: The system revealed that Calcium components have a drastically different radiomic signature compared to Fibrous or IPH components, attaining much higher values across multiple features.
  • Clinical vs. Academic Value: Expert feedback highlighted that while these visualizations are powerful for academic research and "broadening horizons," they require large-scale clinical trials before they can be used for direct bedside diagnosis.

Key Results

The system's multiclass SVM achieved high classification performance for most plaque types, with Area Under the Curve (AUC) scores of 0.98 for Calcium, 0.97 for Fibrous, and 0.97 for IPH. The IPH_lipid category was more challenging, with an AUC of 0.68.

Fig. 3. Performing Case Study 1 with the VA system.Here we show the Plaque size vs Age(A), Plaque Size vs BMI (B), Plaque Composition vs Gender (C), Plaque Size vs BNP (D), Plaque Size vs Troponin (E), Plaque Composition vs Chronic (F).

Through ANOVA analysis, the system successfully identified significant associations between chronic diseases and radiomic features. For instance, the relationship between smoking and the orig_glrlm_RLNU feature yielded a P-value of 0.0001, proving a highly significant correlation that was easily discoverable through the system's interactive interface.

Limitations and Future Work

The authors acknowledge several constraints:

  • Data Scarcity: The small sample size (22 subjects) limits the breadth of the findings and the variety of visualizations that can be effectively employed.
  • Lack of Temporal Data: The current system provides a snapshot in time. Future iterations aim to include longitudinal data to track how plaques change before and after treatment.
  • Visual Overlap: In complex views, color overlap can make interpretation difficult—a common challenge in high-dimensional data visualization.
  • Expanded Indicators: Experts suggested adding more biochemical data, such as lipid levels and white blood cell counts, to further enrich the analysis.

Summary updated

Abstract

Carotid artery plaques can cause arterial vascular diseases such as stroke and myocardial infarction, posing a severe threat to human life. However, the current clinical examination mainly relies on a direct assessment by physicians of patients' clinical indicators and medical images, lacking an integrated visualization tool for analyzing the influencing factors and composition of carotid artery plaques. We have designed an intelligent carotid artery plaque visual analysis system for vascular surgery experts to comprehensively analyze the clinical physiological and imaging indicators of carotid artery diseases. The system mainly includes two functions: First, it displays the correlation between carotid artery plaque and various factors through a series of information visualization methods and integrates the analysis of patient physiological indicator data. Second, it enhances the interface guidance analysis of the inherent correlation between the components of carotid artery plaque through machine learning and displays the spatial distribution of the plaque on medical images. Additionally, we conducted two case studies on carotid artery plaques using real data obtained from a hospital, and the results indicate that our designed carotid analysis system can effectively provide clinical diagnosis and treatment guidance for vascular surgeons.
BibTeX
@article{xu2024integrated,
  author = {Chaoqing Xu and Zhentao Zheng and Yiting Fu and Baofeng Chang and Legao Chen and Minghui Wu and Mingli Song and Jinsong Jiang},
  title = {{An Integrated Visual Analytics System for Studying Clinical Carotid Artery Plaques}},
  journal = {J. Vis.},
  year = {2024},
  doi = {10.1007/s12650-024-00983-1},
}
Figure