Page 354 - IJB-9-4

International Journal of Bioprinting Bioprinting with machine learning

Figure 8. Image pre-processing during defect detection of extrusion-based bioprinting. Reprinted from ref. [90] under the terms of the Creative Commons
CC-BY license.

effectively detect quality errors during 3D bioprinting, and the research developments of machine learning in the
a bigger error signified that a higher detection accuracy field of 3D bioprinting in recent years, hoping to promote
was reached. Tebon et al. utilized a conventional neural the combination of the two technologies further. First,
[91]
network to detect high-throughput drug screening problems the basic principles of k-nearest neighbor, long short-
in 3D-bioprinted organs. A quantitative phase imaging term memory, and ensemble learning are introduced.
technology was employed to obtain relevant images. The Then, the application of machine learning in additive
training dataset contained 100 manually marked organ- manufacturing is reviewed. The in-depth analysis of
like objects in randomly selected imaging frames. The team additive manufacturing technology is helpful to study 3D
demonstrated that the detection effect of the method was bioprinting, which is essentially an additive manufacturing
good, providing a new solution for the rapid screening of technology. Finally, the existing work of machine learning
3D-bioprinted organs. Tröndle et al. obtained a renal in bioink preparation, printing parameter optimization,
[92]
sphere by bioprinting and tested its toxicity based on deep and printing defect detection of 3D bioprinting are
learning technology. Due to the precise deposition of low- summarized. It is expected that this paper can inspire
volume, low-viscosity bioink, the renal sphere was generated more and better research and development of methods
by drop-on-demand bioprinting. They obtained the relevant concerning the combination of 3D bioprinting and
images by fluorescently labeling toxic substances on the machine learning.
kidney sphere. A hyper-parameter Bayesian-optimized
convolutional neural network evaluated the toxicity of renal Acknowledgments
spheres through image processing. The dataset utilized to None.
train the neural network was created by the researchers
using single cell images. Experiments showed that the Funding
accuracy of the model could reach 78.7%.
This work is funded by the National Research Foundation
7. Conclusion of Korea (No. NRF-2022R1A2C2002799).

The rapid development of 3D bioprinting technology in Conflict of interest
recent years is accompanied by the outstanding outcomes
regarding its application. Although there are still many The authors declare no conflict of interest.
problems to overcome, new attempts have been made.
In recent years, machine learning has been applied in Author contributions
many cases, which gives a solid impetus of incorporating Conceptualization: Sang Woo Joo
machine learning to various fields, and the scope of Writing – original draft: Hongwei Ning
application is also expanding. This paper summarizes
Writing – review & editing: Teng Zhou, Sang Woo Joo

Volume 9 Issue 4 (2023) 346 https://doi.org/10.18063/ijb.739