Page 184 - IJB-7-4

Support-Vector-Machine-Guided Parameter Selection for Extrusion-Based Bioprinting
versatile, and can be used with bioinks having a much printer, such as the height of the nozzle, printing speed,
wider range of viscosities . Extrusion bioprinting can be and extrusion speed. Material parameters are related
[10]
further characterized by two types of printers: pneumatic- to properties of the material being used, such as its
based and displacement-based . In a pneumatic-based composition, viscosity, and storage or loss modulus.
[11]
printer, pressure in the reservoir is manipulated to force To encompass both types of parameters, nozzle gauge
material through the nozzle onto the print bed. They and path height were selected for process parameters,
allow for precise pressure control which is important and composition and nozzle temperature were selected
to maintaining cell-viability during printing. Motor- for material parameters. Nozzle temperature may be
based printers utilize a motor to push material down the considered as falling under both categories due to
reservoir and through the nozzle. Despite less precise thermoresponsive viscosity change of PL 127.
pressure control, motor-based printers provide better Machine learning (ML) is currently the most rapid
spatial, and flow control and are therefore a better choice developing field and has tremendous potential in 3D
when using high viscosity materials such as those in printing in terms of developing materials and processes.
[12]
this study. ML is a tool that establishes statistical models to analyze
The overarching type of material used in underling behaviors of a system and give predictions
biofabrication is the bioink. Bioinks are soft materials based on training data . There have been several studies
[21]
which contain living cells that are essential for trying to fast-track optimal bioprinting parameters
prospective applications [13,14] . A common type of bioink and predict printing outcome based on ML algorithms.
is the hydrogel, which is a highly hydrated cross-linked Conev et al. used Random Forest (RF) classifier and
polymer network capable of providing a tissue-like regressor to identify suitable printing conditions to
environment for cells [13,15] . Because of their high hydration recommend for 3D extrusion printing of poly (propylene
and ability to form 3D structures, hydrogels are an ideal fumarate). The authors trained the two RF models
candidate for bioprinting which allows cells to survive using a previous factorial design datasets and explored
and grow . Thermo-responsive hydrogels are of interest the significance of each parameter based on the feature
[13]
in 3D bioprinting due to the opportunity to manipulate weights . Menon et al. used hierarchical ML (HML)
[22]
their properties through temperature control to assist in to predict 3D printing of silicone elastomer. A physical
printing. Among this group of hydrogels is materials such modeling layer was integrated in the HML framework,
as gelatin, methylcellulose, and PEO-PPO-PEO block and the model was trained on 38 data points. Previously
copolymers (trade named Pluronic) . Pluronic F127 unseen data were discovered by the HML with high print
[11]
(PL 127) is of particular interest due to its success for fidelity and 2.5 times higher printing speed . Ruberu
[23]
uses as a sacrificial material which can be printed along et al. employed Bayesian Optimization (BO) to find the
[3]
with other materials then easily removed or washed away optimal printing parameters for 3D extrusion printing of
leaving other materials intact [11,16] . gelatin methacrylate and methacrylated hyaluronic acid
To characterize outcomes, the term printability composite bioink. The number of experiments required to
is often used. Printability is defined as the geometrical reach global optima ranged up to 47 with different bioink
difference between the designed print and the experimental composition, which was greatly reduced compared with
print . Printability is often characterized quantitatively full factorial design (6000 – 10,000) . However, these
[15]
[24]
(termed print fidelity ) using numerical indices of ML models still require quite large amount of training data
[15]
a print’s dimensions and pores [17-19] or qualitatively points. Considering the cost of biomaterials, living cells,
through visual inspection for tears, breakage, or overall and biofactors, we are looking to adopt an algorithm with
performance. Before testing, printability may also a minimal requirement on the number of training data that
be evaluated on the basis of material properties and can still perform well.
performance [15,20] . For this experiment and PL 127’s The objective of this study is two-fold. The first
applications in bioprinting, manipulating parameters to objective is to evaluate the effects of printing parameters
create a high-fidelity print is desirable. To simplify and on the printability of PL 127 (Figure S1). The previous
apply experimental results, width index is used as the research has been conducted on other thermoresponsive
primary indicator for printability in this experiment. hydrogels, including gelatin and alginate [2,17,19,25] , or
Four parameters were selected for testing based on the has used pneumatic-based printers which allow for
hypothesis that they would have a significant impact on manipulation of pressure which is often studied due to
the width index and therefore printability of PL 127. its significant impact on print outcomes . The first study
[26]
Printing parameters are the wide range of variables focuses on establishing the understanding of how each
which can be adjusted to impact printing outcomes. parameter affects the outcome of extrusion printing of
Printing parameters can be separated into two categories. PL 127. The second objective of this study is to utilize
Process parameters are factors which are set by the Support Vector Machine (SVM) algorithm to select

180 International Journal of Bioprinting (2021)–Volume 7, Issue 4