Page 188 - IJB-7-4

Support-Vector-Machine-Guided Parameter Selection for Extrusion-Based Bioprinting
A B

C D

Figure 2. Index representation of parameters. (A) Effects of path height on width index of PL 127 printing. (B) Effects of nozzle gauge.
(C) Effects of PL 127 concentration. (D) Effects of nozzle temperature during printing. Horizontal line indicates ideal index value of 1,
* denotes statistical significance (P < 0.05). Pictures indicate tests which were unable to form the standard three-layer grid. Baseline
parameters: 0.3 mm, 25G, 30%, 23°C. Scale bar is 1.87 mm. Distance between two dashed lines is 0.4 mm.

A B

Figure 3. Parameter selection and construction of process map based on SVM prediction. (A) Training data points selection based on UD
with three parameters and four levels and 12 tests (U (P )). (B) 3D process map generated by support vector machine classifier libsvm with
4
12
3
regularization parameter C = 10 and Gaussian kernel scale parameter g = 0.1.
3.6. Parameter selection based on SVM higher the color map value, the higher the probability of
Training data and its label are shown in Table 2. There data point can yield good print.
are totally 12 data points with four good prints and 4. Discussion
eight bad prints. In the 3D parameter space, all possible
combinations of the parameters are labeled in Figure 3A 4.1. Effect of printing parameters on width index
which is 4*4*4 = 256 in total. “−1” class data points were
labeled in blue cross, “1” class data points were labeled A common factor between multiple parameters is their
in green circle, and the rest were in red circle. Three-fold connection with the viscosity of PL 127. Viscosity is
[12]
cross validation was conducted on the 12 data points defined as the ratio of shear stress to shear rate . PL 127
and an average 83.3% accuracy score was reached with viscosity is dependent on the formation of micelles, which
[11]
C taking 10 and g taking 0.1. The generated 3D process are a formation of connected individual polymer chains .
map reflects the probability for each data point in the For micelles to form, the material must have reached a
parameter space (concentration, nozzle temperature, and certain concentration (critical micelle concentration)
path height) to be classified as good print (class “1”). The and be above the lower critical solution temperature

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