A practical machine learning approach for predicting the quality of 3D (bio)printed scaffolds

Area of Science:

• Biomaterials Science
• Tissue Engineering
• Artificial Intelligence

Background:

• 3D bioprinting enables precise fabrication of tissue engineering scaffolds.
• Artificial intelligence (AI) offers advanced pattern recognition for complex biological applications.
• A significant barrier to AI in tissue engineering is the scarcity of comprehensive, reliable data.

Purpose of the Study:

• To create one of the most extensive open-source datasets for 3D-printed scaffolds.
• To apply diverse AI techniques for analyzing scaffold properties and predicting biological outcomes.
• To facilitate future research in AI-driven tissue engineering.

Main Methods:

• Compiled a dataset of 1171 3D-printed scaffolds with varied biomaterials, cell lines, and printing conditions.
• Employed unsupervised (KMeans clustering) and supervised learning (XGBoost, Random Forest, Neural Networks) algorithms.
• Tuned hyperparameters of over 40 machine learning and deep learning models to predict cell response, printability, and scaffold quality.

Main Results:

• KMeans clustering identified five distinct scaffold groups.
• Classification algorithms like XGBoost and Random Forest achieved high accuracy and F1 scores.
• A custom-built neural network demonstrated precise predictive capabilities for scaffold characteristics.

Conclusions:

• The developed dataset and AI models provide a robust foundation for advancing 3D bioprinting and tissue engineering.
• Publicly available data and code accelerate AI integration and innovation in the field.
• This work highlights the potential of AI to overcome data limitations and enhance scaffold design and performance.