Statistical and Machine Learning Approaches for Virtual Population Generation in In-Silico Cardiovascular Trials.

Area of Science:

• Biomedical Engineering
• Computational Biology
• Medical Imaging

Background:

• In-silico clinical trials require realistic virtual patient data for evaluating medical devices like cardiovascular stents.
• Traditional methods for generating such data are resource-intensive and may lack diversity.
• The InSilc project aims to develop advanced simulation tools for cardiovascular research.

Purpose of the Study:

• To develop and validate a hybrid methodology for generating high-fidelity virtual human cardiovascular datasets.
• To improve the accuracy and realism of virtual patient populations for in-silico clinical trials.
• To enhance the evaluation of novel cardiovascular devices, such as stents.

Main Methods:

• Integration of a statistical modeling technique (multivariate normal distribution) with a machine learning (ML)-based generative model (Conditional Tabular Generative Adversarial Networks - CTGAN).
• Statistical model addresses missing data and non-positive definite covariance matrices.
• CTGAN synthesizes patient populations while preserving statistical integrity of real-world data.

Main Results:

• The hybrid approach successfully generated a virtual population of 10,000 patients after initial validation.
• The statistical model showed superior accuracy in predicting anatomical parameters.
• The ML approach excelled in capturing complex inter-variable relationships within the dataset.

Conclusions:

• The combined statistical and ML methodology enhances the simulation of diverse patient populations for in-silico trials.
• This approach improves the robustness, efficiency, and cost-effectiveness of clinical trials.
• The methodology advances in-silico clinical trials, potentially improving patient safety and device evaluation.