AI-powered printability evaluation framework for 3D bioprinting using Hausdorff distance metrics

Abstract 3D bioprinting enables rapid fabrication of complex biological structures for tissue engineering applications. However, optimizing bioink formulation remains challenging due to complex relationships among material properties, printability, and cell viability. While the perimeter ratio (Pr) is commonly used to evaluate printability, it cannot adequately capture the full geometric fidelity required for comprehensive printability assessments, thereby limiting robust bioink design. To address this limitation, a novel Hausdorff distance (HD) metric is employed to quantify printability, directly measuring the maximum deviation between the designed and printed structures. Furthermore, multiple machine-learning approaches were applied to alginate-hyaluronic acid composite inks and rat pheochromocytoma-derived PC12 cells to assess printability and cell viability. Rheological parameters were characterized using support vector regression (SVR) with R ² ⩾ 0.974. Multi-layer perceptron (MLP) regressors achieved R ² values of 0.932 and 0.945 when predicting HD values of printed grid structures and cell viability, respectively. A regression-based convolutional neural network (CNN) was developed to predict HD values directly from grid structure images, achieving an R ² of 0.986. Through optimization, optimal as-extruded cell viability (⩾95%) was achieved while maintaining high printability (HD ⩽ 0.20). The optimal ink composition was further demonstrated with good long-term cell viability and proliferation potential. This proposed AI-integrated approach can dramatically reduce ink optimization time by rapidly predicting rheological properties, printability, and cell viability from minimal experimental data.