Machine learning for 3D printed multi-materials tissue-mimicking anatomical models

Abstract

Polyjet, a material jetting 3D printing technique, has been widely used for the fabrication of patient-specific anatomical models owing to the toolless fabrication technique and its ability to print multiple materials in a single part. Although the fabrication of anatomical models with high dimensional accuracy has been demonstrated, 3D printed anatomical models with tissue-mimicking properties have not been realized. In this study, a composite layering design was used to tune the shore hardness and compressive modulus of the Polyjet-printed parts in an attempt to mimic the properties of human tissues. 216 specimens (with 72 combinations of design parameters) were printed and tested to develop the material library for the anatomical models. An analytical model was developed to estimate the effective compressive modulus and shore hardness of the composite laminate. A neural network was used to learn the multi-dimensional relationship between the design parameters and mechanical properties. The 5-33-2 network size is found to be the optimum neural network structure with a mean square error of 0.98% for the compressive modulus, lower than the traditional response surface method model. A genetic algorithm was used to search the design space for the most optimum design parameters for the targeted effective shore hardness.