Page 190 - IJB-7-4

Support-Vector-Machine-Guided Parameter Selection for Extrusion-Based Bioprinting
material would cut across rather than form a proper printers. To confirm that this effect did not impact testing,
corner . Low path heights can have the opposite affect a short second experiment was conducted. 21, 23, 25,
[25]
due to the nozzle interfering and spreading the material 27, and 30G nozzles were used with a room temperature
out into over-deposited thick lines (Figure 4B). sample of 30% PL 127 and extruded for 2 min each
Nozzle gauge was not found to have a significant (Figure 2S). The amount of material was then weighed
impact on width index, but some other differences were and compared, revealing that all nozzles did extrude the
observed between tests. A 21G nozzle corresponds to an same amount of material. Additional testing revealed
inner diameter of 0.58 mm and an outer diameter of 0.81 mm, that effects are seen at nozzle gauges higher than 27G
23G to an inner/outer diameter of 0.43 and 0.635 mm, such as 30G (Figure 2B).
25G to an inner/outer diameter of 0.3 and 0.5 mm, and
27G to an inner/outer diameter of 0.2 and 0.4 mm. It is 4.2. Optimal parameter selection based on SVM
expected that a change in nozzle diameter will not affect We selected two locations on the parameter space; one
line width as the flow rate out of the nozzle should not be having higher than 75% probability, the other having lower
affected if the bioink is considered incompressible. With than 25% probability (Figure 5). The scaffold printed with
all other parameters held constant (particularly print speed the parameters from low probability region cannot form
and extrusion speed) the flow rate out of the nozzle is also continuous and stable structure and has a low printability.
constant, and line width therefore cannot change since the The printed cube and grid 3D structure were not able to
same amount of material is being deposited. A difference form uniform and accurate shape as desired (Figure 5A).
was noted in the second layer performance of tests. When While the scaffold printed with the parameter from high
using wider nozzles, the second layer is stretched more printability region was able to generate high printability
and may tear or fail completely. This can lead to thinner stable scaffold with multiple test prints having width index
first-layer width as surface tension and the weight of evaluated at 0.998 ± 0.049 (mean ± standard deviation).
the second layer are absent, leading to less spreading in The printed cube and pyramid structure maintained good
single-layer prints. 21G tests were unable to form a second fidelity and uniformity (Figure 5B).
layer, 23G tests could form a second layer some errors, There exists a complex interplay between various
and 25G and 27G were able to form a consistent second printing parameters to achieve desired printability of the
layer. This problem necessitated single-layer prints for an scaffold. The impact on the scaffold printability caused
accurate comparison of how nozzle gauge effects purely by changing one parameter can always be compensated
line width. However, a complete approach to defining a by adjusting another. For example, when printing with
“best” nozzle gauge would require these problems with a low concentration of PL 127, the low viscosity of
certain nozzles be considered. the material could be compensated for with a high
Effects could be seen if nozzle diameter is small printing temperature which would increase viscosity.
enough to cause a large pressure buildup inside the nozzle Understanding these relations creates the possibility of
tip resulting in a push back on the printer motor. Pressure any number of “best” parameter combinations which
effects would then impede motor function and extrusion create high fidelity prints. The SVM process optimization
speed would be effectively lowered. To describe this issue method provides a solution to analyze the sophisticated
a mathematical model of flow rate in the nozzle tip (Eq. 3D bioprinting black box. Using a minimal preselected
5) can be examined : training data points can assist construct SVM prediction
[33]
on a volumetric parameter space so that the optimal
 n n n − 1   / ∂P z  ∂ 3 +n 1 printing parameter combinations can be acquired directly
=
Q   3 + 1 n   0   2   R n (5) without tedious trial and error experiments.
0
We only used three parameters that were
where Q represents flow rate, n is the power law hypothesized to have a significant impact on printability.
index of the fluid, γ is the shear rate, P is pressure, z is In fact, a plethora of parameters, such as blend ratio
the direction in the nozzle axis, η is the limited viscosity, (composite bioink), extrusion pressure (for cell
0
and R is the nozzle radius. If the flow rate (Q) out of the encapsulated printing), and crosslinking strategies
nozzle is to remain constant while the nozzle gauge (R) (e.g. duration and timing), should also be included. In
decreases, then pressure (P) must increase to balance the addition, utilizing governing equations to make more
equation as no other variables will change. In smaller physically informed choice on the parameters is also
nozzles, this pressure increase could be high enough promising to build a more generalized model.
as to unintentionally lower extrusion speed because of There were various quantification methods reported
push-back. This is a possible drawback of motor-based on the printability of a scaffold, which significantly
printers which is avoided with pressure-based pneumatic affects the generalization of ML model since the label

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