Machine-Learning-Assisted Quantitative Printability Assessment in Extrusion-Based Bioprinting—A Systems-Engineering Proof-of-Concept

Abstract

Extrusion-based bioprinting is governed by coupled material, extrusion, and motion parameters, yet printability is often assessed using isolated rheological tests or qualitative geometric inspection. This study developed a systems-engineering framework for the quantitative assessment of non-cellular syringe-extrusion printing, with relevance to future bioprinting applications. Thirty-six constructs covered a complete 3 × 3 × 4 factorial design comprising three nozzle diameters, three printhead velocities, and four reference trajectories, with one independently printed construct per unique condition. Fiji/ImageJ analysis quantified filament width, edge roughness, curvature, and trajectory fidelity, and a study-relative Printability Score (PS) integrated four normalized geometric error domains. PS rankings were robust to moderate changes in component weighting (Spearman ρ = 0.955–0.999). PCA identified distinct deposition- and geometry-related modes; PC1, PC2, and PC3 explained 51.18%, 23.06%, and 11.69% of the variance, respectively (85.92% cumulative). Under LOOCV, raw-input Ridge Regression achieved R2 = 0.713, MAE = 0.195, and RMSE = 0.285, whereas Gradient Boosting achieved R2 = 0.709, MAE = 0.216, and RMSE = 0.288. Cross-validated permutation analyses identified printhead speed and trajectory geometry as the most informative raw predictors. These findings establish an offline engineering proof of concept rather than replicated confirmatory validation, biological validation, or closed-loop control.