Machine Learning in Predicting Printable Biomaterial Formulations for Direct Ink Writing
Hongyi Chen1,2, Yuanchang Liu1,1, Stavroula Balabani1,3
1Department of Mechanical Engineering, University College London, London, UK.
Research (Washington, D.C.)
|July 20, 2023
Summary
Artificial intelligence, including machine learning algorithms, can predict the printability of 3D biomaterials. This approach accelerates the development of custom 3D printed biomedical devices, reducing trial-and-error methods.
Area of Science:
- Biomedical Engineering
- Materials Science
- Computational Biology
Background:
- Three-dimensional (3D) printing offers patient-specific customization in biomedical engineering.
- Current biomaterial ink development relies on time-consuming trial-and-error methods.
- Expert knowledge is often required to navigate complex formulation variables.
Purpose of the Study:
- To develop predictive models for biomaterial printability using machine learning (ML).
- To accelerate and optimize the development of 3D printable biomaterial formulations.
- To guide the selection of materials and solvents for specific 3D printing applications.
Main Methods:
- Constructed and evaluated ML algorithms: decision tree, random forest (RF), and deep learning (DL).
- 3D printed 210 formulations comprising 16 bioactive/smart materials and 4 solvents.
- Assessed printability of each formulation and trained ML models on experimental data.
Main Results:
- All ML algorithms successfully predicted biomaterial ink printability.
- The RF algorithm demonstrated the highest accuracy (88.1%), precision (90.6%), and F1 score (87.0%).
- Deep learning generated a printability map with finer granularity, aiding ink development.
Conclusions:
- ML provides an effective and novel strategy for developing 3D printable biomaterials.
- Predictive modeling significantly enhances the efficiency of biomaterial ink formulation.
- This approach supports the advancement of custom 3D printed biomedical applications.
