An ink-insensitive deep learning model for improving the printing quality in extrusion-based bioprinting

Extrusion-based bioprinting (EBB) is an efficient, simple, and cost-effective bioprinting technique. However, because of the high complexity of biomaterial inks, a trial-and-error approach is typically required for determining the optimal printing parameter configurations. Thus, the quality and reproducibility of printed scaffolds remain a concern. In this study, we integrated flow sensing into the EBB and replaced conventional fixed-value printing parameters by capturing the time-series data of all printing parameters to enhance the monitoring of the printing process. To improve the EBB adaptability, we conducted experiments with three biomaterial inks with distinct properties under various parameter configurations to achieve ink-insensitive linewidth prediction through the construction and assessment of deep learning models. Additionally, two explainable artificial intelligence methods were used to analyze the decision-making process of the deep learning model. This analysis not only enhanced model reliability but also identified key features in printing parameters and time-series data. These results can be used to improve the efficiency and quality of the EBB processes by using various biomaterial inks.