Prediction of Left Ventricular Mechanics Using Machine Learning.

Area of Science:

• Computational mechanics
• Biomedical engineering
• Machine learning applications

Background:

• Accurate simulation of left ventricular (LV) mechanics is crucial for understanding cardiac function and disease.
• Traditional finite element (FE) methods, while accurate, are computationally intensive and slow for real-time applications.
• Machine learning (ML) offers a potential solution for accelerating these complex biomechanical simulations.

Purpose of the Study:

• To develop and validate a real-time left ventricular (LV) mechanics simulator utilizing machine learning (ML).
• To compare the predictive accuracy and speed of ML models against traditional FE simulations for LV mechanics.
• To assess the capability of ML in capturing complex LV pressure-volume dynamics and stress distributions.

Main Methods:

• Generation of a training dataset using FE simulations of the LV with a hyperelastic fiber-reinforced material model.
• Inclusion of active (myofiber contraction) and passive (myocardium behavior) properties in the LV constitutive equation.
• Application of eXtreme Gradient Boosting (XGBoost) and Cubist ML algorithms for predicting LV pressures, volumes, and stresses.

Main Results:

• ML models demonstrated comparable accuracy to FE computations in predicting LV pressure and volume.
• The ML approach successfully captured the characteristic shapes of LV pressure and pressure-volume loops.
• ML predictions were significantly faster than FE simulations, with reported mean absolute errors for volume, pressure, and stress metrics.

Conclusions:

• Machine learning provides a computationally efficient alternative to FE methods for simulating LV mechanics.
• The developed ML model can serve as a valuable tool for rapid prediction of LV behavior.
• Further training of the ML model on diverse subject data can enhance its real-world applicability and predictive power.