Leveraging transfer learning for efficient bioprinting

Bioprinting is a promising family of processes combining 3D printing with life sciences, offering the potential to significantly advance various applications. Despite numerous research efforts aimed at enhancing process modeling, optimizing capabilities, and exploring new conditions, there remains a critical need to enhance process efficiency. Experimental data are paramount for improving models. Nevertheless, it is practically unfeasible to explore a multitude of conditions (e.g. different material formulations, process parameters, machines, setups), especially given the experimental constraints of budget and time. Leveraged by in-situ bioprinting monitoring, this paper explores a set of transfer learning (TL) methods designed for resource-efficient bioprinting modeling, aiming to merge established knowledge with new experimental conditions. TL encompasses machine learning strategies focused on transferring knowledge across distinct, yet similar, domains. TL is applied to an extrusion-based bioprinting case study for printability response modeling. The knowledge acquired from a model trained on one material (the source) is transferred to a new material (the target), under conditions of limited experimental data availability. Eventually, the accuracy of the transferred model is assessed and compared against a reference no-transfer scenario, which is developed from scratch following conventional practices. Furthermore, giving high importance to the experimental effort reduction, a sensitivity analysis altering the number of experimental training points is performed to assess performances and limitations of the method. This method demonstrates the feasibility of knowledge transfer in bioprinting as a catalyst for more sophisticated applications across diverse printing conditions, materials, and technologies to advancing this technology towards achieving its full potential.