Page 62 - IJB-9-4

International Journal of Bioprinting Machine learning and 3D bioprinting

of the developed model for a wide range of material knowledge, experiments, and empirical experience. It
properties was questionable. Hence, fine-tuning of the also provides a simulation tool to model the impact of
optimized printing parameters may be utilized to further the printing parameters on the filament/droplet diameter,
improve the bioprinting performance. fiber quality, shape fidelity, and layer stacking. Although
In addition to printing resolution, bioprinting often the current models are developed under a controllable
requires good regularity in EBB layer stacking. Two of range of material properties, they may suffer from poor
these can be incorporated into one scoring metric for generalization and robustness under varied material
printing quality evaluation. To optimize this metric, properties or modified experimental protocols.
Bayesian optimization (BO) has been used to explore 2.3. ML in biomaterial/bioink optimization
the printing parameter space for various bioinks . The An increase in cell culture applications in tissue engineering
[30]
inputs of this BO model consisted of gelatin methacryloyl and drug screening has resulted in a growing demand
(GelMA) inks with three concentrations, three inks with for biomaterial/bioinks [14,16] . Higher concentrations of
mixed concentrations of GelMA and hyaluronic acid bioink/biomaterial may yield good shape fidelity, but poor
methacrylate (HAMA), and printing parameters (bioink printability during extrusion.
reservoir temperature, extrusion pressure, print-head
speed, and platform temperature). Compared to trial-and- Increasing the viscosity of biomaterial/bioink can
error experiments, this ML application can systematically improve shape fidelity, but may compromise printability.
accelerate the retrieval speed of printing parameters and Because traditional governing equations cannot effectively
bioinks for high-quality printing. and efficiently handle such bioink/biomaterial optimization
tasks, manual calibration with numerous experiments was
Traditional ML methods have also been applied to conducted.
optimize process parameters for EHD inkjet printing To support a new material design workflow, several
and electrospinning. For example, statistical regression ML methods have been implemented to optimize the
analysis, neural networks (NNs) trained with genetic composition and viscosity of biomaterial/bioink to
algorithms (GA-NN), and backpropagation NNs have digitally manipulate printability and shape fidelity [40-42] . For
been compared when predicting droplet diameter in EHD example, inductive logic programming has been applied
inkjet printing . The standoff height, applied voltage, and to investigate the rheological properties of hydrogel inks
[28]
ink flow rate were the inputs to the prediction models. and their corresponding printing qualities . Rheological
[32]
The GA-NN model outperformed the other two models properties, such as elastic modulus and yield stress, were
in most cases when predicting droplet diameter. Similarly, classified into three classes, and extrusion capability and
multiple regression (MLR), multilayer perceptron shape fidelity were classified into two classes. The analysis
artificial neural network (ANN), and SVM models have results from the ML models indicate that printable
been used to predict the electrospun diameter of PCL/Gt ink should have a high elastic modulus for high shape
nanofibers . With the key input parameters identified by fidelity and low yield stress for extrusion. Based on this,
[24]
saliency analysis, the ANN model demonstrated the best a multiple regression model was proposed to quantify the
performance compared with other models [24,39] .
relationship between ink formulations and printability.
Theoretically, the printing process parameters usually Another regression model, hierarchical machine learning
have conflicting multiperformance characteristics. For (HML), was reported to determine the material properties
example, a higher applied voltage may increase the of sodium alginate prints with high/low fidelity .
[33]
biomaterial/bioink printability but lower the printing The middle layer in the HML was constructed based
resolution. A statistical-based method named “desirability on acquired knowledge of the flow-gelation process
function analysis” has been reported to simplify of alginate. The dimensional similarity between the
the process parameter optimization in multicriteria deposited structure and original design was used to
objectives . Furthermore, this statistical method that evaluate the HML model. This method can effectively
[31]
relies on composite desirability may not be able to handle guide high-fidelity EBB scaffold fabrication by optimizing
the delicate relationship between the process parameters in biomaterial formulation and printing parameters. This
EHD inkjet printing. may be a promising way to scale up the biomaterial/
As discussed above, traditional ML algorithms have bioink shape fidelity study through the combination of a
been used to rank input feature relevance, establish small number of iterative tests, practical experience, and
printing performance models, and optimize the process theoretical knowledge.
parameters. This may simplify the process–material– In addition, hydrogel inks may require crosslinking
performance model building with limited in-depth to maintain the shape and retain the printed structures

Volume 9 Issue 4 (2023) 54 https://doi.org/10.18063/ijb.717