• clinical translation
• quality control
• bioprinting
• tissue engineering

Bioprinting has seen an exponential increase of attention in recent years from both academia and industry as a promising solution to fabricate constructs for Tissue Engineering applications, including implantation, in vitro models, and drug screening. Even though important results have been shown in literature, the field is currently suffering from a severe lack of clinical translation examples, mainly due to both technological limitations (e.g., need for new inks, need for advancements in current fabrication technologies, incorporation of vasculature inside the construct) and regulatory barriers (e.g., definition of the classification for the bioprinted product, development of bioprinting-specific standards).
Considering these challenges, a major need for the bioprinting field is the implementation of quality control strategies to reduce inter-batch variability, guarantee that the final product is close to the designed model, and comply to relevant quality-related international standards.
In this context, the objective of the PhD work is the formulation and development of quality control strategies for different bioprinting processes, which will be applied for the development and fabrication of a case study Advanced Therapy Medicinal Product (ATMP) device in the context of the European H2020 project GIOTTO (GA: 814410).
The starting point of the doctoral project was the design and fabrication of a custom-made bioprinter, which served as the basis for the development of add-on quality control systems. In particular, the bioprinter is characterized by a high precision (5 µm) and repeatability (± 10 µm of bidirectional repeatability) in positioning thanks to the use of stepper-actuated linear stages. The robust positioning system guarantees that no errors are introduced in the scaffold production due to incorrect movements of the machine. From a software point of view, the bioprinter employs LinuxCNC, an open-source control software for numerical control machines (e.g., lathes, cutters). LinuxCNC can enable real-time control of the bioprinter positioning system, as well as the integration of new modules thanks to its software architecture. With this system, at any given time two tool-heads are in place in the bioprinter for multimaterial and multiscale fabrication. The user can select from three interchangeable deposition tool heads that can be swapped in place depending on the application: i) piston-actuated extruder, designed for highly viscous solutions; ii) thermal drop-on-demand inkjet, to pattern the scaffold with picoliter droplets of liquid material; iii) fused deposition modelling (FDM) extruder, to introduce rigid polymeric support inside the scaffold.
For each tool-head, different strategies were implemented to assure consistent deposition and control over the printing process. In particular, the extruder stepper actuator was equipped with a rotary encoder that monitors the position in closed-loop feedback to compensate for under- or over-extrusion artifacts. A custom-made electronic control system was implemented for the inkjet tool-head to modify the operating voltage and so enable the patterning of custom solutions. Finally, a sensor was designed and fabricated to monitor the filament diameter for the FDM system, with the aim of keeping a constant flowrate by compensating for variation of the diameter.
Then, advanced quality control solutions were developed specifically for the extrusion tool-head, since currently it represents the most used manufacturing technology in bioprinting. Firstly, an analytical model of the extrusion bioprinting process was formulated to: i) predict the printability of a given biomaterial ink using a specific bioprinting apparatus, scaffold geometry, and set of printing parameters, and ii) if the biomaterial ink is found to be printable, provide a set of optimized printing parameters to be used for experimentation. The model was developed by taking into consideration multiple aspects of the process (referred to as stages for brevity), including:
1. the extrudability, i.e., the easiness of extrusion of the given material from the syringe needle;
2. the line deposition process when printing the first layer of the scaffold;
3. the stabilization of a three-dimensional woodpile scaffold after printing to evaluate the shape fidelity of the final product.
For each model stage, relevant equations were formulated by considering different constitutive equations for the biomaterial ink (i.e., Newtonian fluid, power law, Herschel Bulkley), and the equations were experimentally validated. The analytical model was then implemented in both a standalone program and web-based application to help the bioprinting scholars determine a priori the ink printability and optimize the printing parameters. Concurrently, to support the model use, a set of rheological characterization experiments (based on recognized material characterization standards) were defined to find the important material properties for the model.
Building on the aforementioned results, a novel, artificial intelligence-based solution for the in-process monitoring and automatic parameter optimization was developed. Briefly, a comprehensive dataset of multiple scenarios was constructed by recording the printing process from a front view using a high-resolution webcam. Each video corresponded to a print with a different combination of relevant parameters, including layer height, flow, printing set-up (i.e., pneumatic and piston-actuated extrusion), material color. Two main errors were introduced in the dataset resulting from a non-optimal printing parameters combination, including under- (i.e., not enough material is extruded) and over-extrusion (i.e., too much material is extruded). After sampling and frame pre-processing, the resulting dataset was used to comprehensively optimize an ad-hoc convolutional neural network by considering as main requirements the high classification accuracy (around 94%) and the fast response time (around 180 ms to process 30 frames on CPU). The model was used for monitoring the printing process online to stop the print if an error occurred before completion, to save time and reduce material waste (at least 20% of the material for a print saved). Furthermore, an automatic parameter optimization system based on a series of consecutive prints with varying parameters was developed to optimize the parameters automatically, without the need for user intervention and material characterization. Altogether, the two strategies represent a comprehensive software solution for quality control of the extrusion bioprinting process.
Even though quality control during manufacturing is a key requirement for clinical translation of the bioprinted product, it is also important to standardize not only the production process but also the nomenclature related to the field, to facilitate the development of shared libraries of materials, protocols, and bioprinting-specific standards. In this context, during this PhD project a novel application of Natural Language Processing (NLP) to the bioprinting literature was developed, with the goal of extracting knowledge from scientific papers. In particular, the approach is based on two main data sources, the abstracts and related author keywords, which are used to train a composite NLP model. This is based on: i) an embeddings part, which generates word vectors (i.e., dense numerical representations of a word) for an input keyword, and ii) a classifier part, to label it based on its word vector into a manufacturing technique, used material, or application of the bioprinted product. The composite model was trained and optimized in a two-stage optimization procedure to yield the best classification performance (around 90% accuracy). The annotated author keywords were then found in the abstract collection to both generate a lexicon of the bioprinting field and extract relevant information, like technology trends and the relationship between manufacturing-material-application. The proposed approach can serve as a basis for more complex NLP-related studies toward the automated analysis and standardization of the bioprinting literature.
Finally, the bioprinter platform alongside the developed quality control solutions were applied for the fabrication of an ATMP device in the context of the GIOTTO project. The Device is a porous, graded structure, produced through bioprinting technologies (i..e, FDM, inkjet), which can direct the diffusion of relevant molecules towards an osteoporotic femoral fracture and so accelerate fracture healing. During the project multiple deposition technologies were used to fabricate prototypes and scaffolds at different stages of the device development. The experiments started with the assessment of the printability of different filament compositions using FDM and the determination of the optimal printing parameters for each one. The optimized set of printing parameters were used for FDM production of scaffolds (i.e., porous cylinders, 5 mm diameter, 1 mm height) for in vitro testing. The scaffold quality was characterized after printing in terms of their dimensional accuracy and overall porosity when compared to the original design. An open-mould method was developed for the FDM fabrication of scaffolds (i.e., non-porous cylinders, 2 mm diameter, 0.2 mm height) for in vivo testing. Post-printing quality assessment was performed by measuring their dimensions and evaluating their shape at an optical microscope. The results showed that the developed method could yield tight dimensional tolerances (i.e., ± 0.1 mm in both diameter and height) according to the fabrication constraints. Moreover, superparamagnetic iron oxide nano-particles (SPIONs) solutions printability was assessed through inkjet printing. The experiments helped determine the best performing solvent by evaluating the printed solutions at an optical microscope and looking at the stability of the printing process through subsequent prints. Finally, a combination of FDM and inkjet was used to demonstrate the printer ability to fabricate multimaterial and multiscale scaffolds.
All in all, these experiments served as validation of both the developed bioprinting platform and the quality control strategies, showing how these solutions can be used to obtain repeatable, high quality results and so can be envisioned to move the bioprinting field to more impactful clinical applications.