Perfect prosthetic heart valve: generative design with machine learning, modeling, and optimization.

Area of Science:

• Biomedical Engineering
• Computational Science
• Medical Device Design

Background:

• Traditional medical device design relies on computer-aided design (CAD) and finite element method (FEM), which are time-consuming and limit design exploration.
• Optimizing medical device geometry is crucial for performance and patient outcomes but faces computational challenges.

Purpose of the Study:

• To develop and evaluate a novel generative design approach combining machine learning (ML) and optimization algorithms for efficient medical device geometry optimization.
• To accelerate the design process and identify optimal designs within specified constraints for devices like prosthetic heart valves (PHVs).

Main Methods:

• Evaluated eight ML methods (e.g., neural networks, ensembles) and six optimization algorithms (e.g., Tree-structured Parzen Estimator, Nondominated Sorting Genetic Algorithm).
• Applied the generative approach to design a prosthetic heart valve, using design constraints as inputs.
• Assessed design effectiveness using a scoring system and prediction error rates.

Main Results:

• The combination of ensemble ML methods with Tree-structured Parzen Estimator or Nondominated Sorting Genetic Algorithm proved most effective.
• Achieved Mean Absolute Percentage Errors of 11.8% for lumen and 10.2% for peak stress prediction.
• Optimized designs achieved effectiveness scores of approximately 95%.

Conclusions:

• The proposed generative design approach significantly accelerates the design and optimization of medical devices compared to CAD-FEM methods.
• This ML-driven approach offers a powerful tool for discovering efficient geometries within defined constraints, paving the way for improved medical device development.
• The study provides a publicly available repository with code, datasets, and models to facilitate further research and application.