Page 66 - IJB-9-4

International Journal of Bioprinting Machine learning and 3D bioprinting

papers by searching the Web of Science database. A selection and the subsequent crosslinking at different
nuclei segmentation dataset, Scaffold-A549, was built by degrees to build biological constructs with desired shape
collecting human lung cancer cell images for cell–scaffold fidelity.
interaction studies. This dataset consisted of 20 high- 4.3. Unlabeled data in cell performance analysis
resolution unlabeled CLSM images and one fully labeled The application of supervised ML methods relies on a
image with approximately 800 segmented nuclei. Although carefully and precisely labeled dataset. However, it is
data augmentation can artificially enlarge a dataset by difficult to label nuclei or cells on bioprinted constructs
adding noise and interpolating or extrapolating between using biological images. First, several overlapping cells or
samples in a feature space, its application in bioprinting thousands of nuclei with irregular or deformed shapes are
has not yet been reported.
observed. Varied cell shapes, phenotypes, and types cause
To advance bioprinting using ML, it is essential to this labeling task to worsen. Moreover, a single-cell type
make additional experimental data publicly available may exhibit a wide range of morphologies in terms of
or accessible upon request. Using similar open-source adhesion, proliferation, and migration. Thus, the interest
software for bioprinters and establishing a worldwide in applying unsupervised ML methods to model such
data-sharing network could be promising . However, complex unlabeled datasets has continued to grow. Such
[46]
researchers may not wish to publish the collected data studies aim to automatically identify cell types or behaviors,
because of expenses related to the materials, manpower, assist in the discovery of biological phenomena at the
and facilities. This has impeded the advancement of ML single cell level, and upgrade cell laboratory workflow.
applications for bioprinting.
4.4. Digitalized in situ performance analysis
4.2. Information integration and coordination from The bioprinted constructs are biologically evaluated based
multiple sources on the overall performance of comparative samples. This
Apart from dataset quality, information integration from gap has motivated researchers to explore in situ evaluation
multiple formats and sources is an ongoing research methods. As the biological images of these constructs
issue. In current studies, the collected numerical data or are collected, DL can digitalize them by segmenting cell/
images in bioprinting are used individually for specific nuclei, classifying cell types, and identifying phenotypes.
prediction tasks. For example, features extracted from This is valuable when analyzing bioprinted constructs
the process parameters or material properties can be with multilayer heterogeneous or gradient structures.
conveniently used to develop prediction models [20,39,54] . Currently, overall performance evaluation cannot provide
However, it is difficult to deliver a prediction task when in situ information about the biological response of a
both image and numerical data are input. They are structure. The digitalized cell distribution can also present
equally important and should be synchronously used to an intuitive biological response to the structural design,
develop process–material–performance models or refine layer-specific fiber diameter and pore size, and biomaterial
bioprinted construct designs for better cell performance. composition. Consequently, the cell number and spatial
Further strategic studies should be conducted to integrate behavior of bioprinted constructs can be subjectively and
the diverse data formats when developing ML prediction quantitatively evaluated. This would impel researchers
models. In addition, the emerging technologies, such to combine diverse materials and structures and
as big data and digital twin, may also contribute to the support functional scaffold design with better biological
information integration and coordination. One way is performance and revolutionize construct-based organ or
to build ML training databases using big data curation, disease model studies. More studies using ML to predict
and the other is to build digital twins of human tissue/ tissue formation and create organoid and tumoroid models
organs with cellular resolution and properties . As such, are expected, although these topics seem significantly
[17]
a standard bioprinting simulation practice is expected to challenging.
balance virtual and physical experiments and maximize 5. Conclusion
bioprinting resource utilization.
Bioprinting has demonstrated its ability to produce
Information integration and coordination is of
increasing interest in multi-material printing. To achieve constructs for cell culture in tissue engineering and drug
screening. These applications are still primitive because of
a cost-effective and time-efficient printing, ML is expected the limited fundamental research on process–material–
to integrate and coordinate diverse information in this performance.
process so as to enhance printing resolution, printing path
planning, G-code error detection, and structural stability In this review, the current status, challenges, and
of bioprinted construct. ML can also facilitate the material outlook of ML applications in bioprinting are discussed.

Volume 9 Issue 4 (2023) 58 https://doi.org/10.18063/ijb.717