Using intravascular ultrasound image-based fluid-structure interaction models and machine learning methods to predict...

Area of Science:

• Cardiovascular Research
• Biomedical Engineering
• Medical Imaging

Background:

• Plaque vulnerability prediction is vital in cardiovascular research.
• Intravascular ultrasound (IVUS) provides in vivo data for plaque analysis.
• Understanding plaque biomechanics aids in predicting cardiovascular events.

Purpose of the Study:

• To develop and compare models for predicting changes in plaque vulnerability.
• To identify key predictors of plaque vulnerability using baseline data.
• To evaluate the accuracy of machine learning methods in plaque vulnerability assessment.

Main Methods:

• Acquired in vivo IVUS coronary plaque data from nine patients.
• Constructed fluid-structure interaction models to assess plaque biomechanics.
• Defined a Morphological Plaque Vulnerability Index (MPVI).
• Employed Generalized Linear Mixed Regression Model (GLMM), Support Vector Machine (SVM), and Random Forest (RF) for prediction.
• Utilized ten baseline risk factors to predict MPVI change (ΔMPVI).

Main Results:

• Random Forest (RF) achieved the highest prediction accuracy (91.47%) using a combination of mean wall thickness, lumen area, plaque area, critical plaque wall stress, and MPVI.
• RF improved prediction accuracy by 5.91% compared to GLMM.
• MPVI was the strongest single predictor for GLMM (82.09%) and RF (78.53%).
• Plaque area was the best single predictor for SVM (81.29%).

Conclusions:

• Machine learning models, particularly RF, demonstrate high accuracy in predicting coronary plaque vulnerability changes.
• Biomechanical factors derived from IVUS data are significant predictors of plaque vulnerability.
• MPVI and plaque area are important indicators for assessing plaque instability.