Page 151 - v11i4

International Journal of Bioprinting AI for sustainable bioprinting

on 1050 labeled data points collected from simulations and the diversity and scale of training datasets, including
experiments using four cell lines (fibroblast, stem, cancer, parameters such as crosslinking kinetics and biological
and endothelial cells). Input features included wall shear performance, and developing real-time, adaptive control
stress (1.0–5.0 kPa) and exposure time (100–700 ms), while systems could further enhance the robustness and utility
the output target was cell viability. Model performance was of AI in bioprinting.
evaluated using 20-fold cross-validation, yielding high
predictive accuracy with R values ranging from 0.866 to 3.4. Artificial intelligence-assisted
2
0.964 across cell types. intelligent printing
Beyond material development, bioink formulation, and
Rojek et al. presented an AI-driven approach to parameter optimization, AI-assisted intelligent printing
90
optimize 3D printing efficiency and reduce material waste represents a major step toward adaptive and autonomous
by training artificial neural networks on a dataset of 238 bioprinting systems. These systems leverage advanced
input parameters and eight output metrics, including algorithms, real-time monitoring, and process automation
filament usage, cost, and print time, using experimental to enhance control, precision, and functionality. Advanced
data from 3D-printed elbow exoskeleton components. The algorithms enable in situ bioprinting with acoustic
artificial neural network model (multilayer perceptron levitation, allowing voxel-by-voxel and contact-free
regressor-142-102-8) achieved strong predictive material placement on diverse substrates and expanding
performance (mean squared error = 0.007; testing quality the possibilities of bioprinting. 92,93 AI-powered computer
= 0.9132), allowing a 30-fold reduction in waste and vision systems, feedback loops, and predictive modeling
enabling one free print for every 6.67 prints. have been employed to detect and correct print defects in
Wu and Xu developed a data-driven ensemble real time, improving print success rates. 94
91
learning approach to predict droplet velocity and Chen et al. developed an intelligent printing
92
volume in inkjet-based bioprinting. They conducted system, AcoustoFab, which combines a phased array of
a full factorial design with 243 experiments varying transducers and advanced control algorithms, allowing
polymer concentration, voltage, dwell time, and rise time, omnidirectional and multi-material in situ bioprinting
training predictive models (random forest, least absolute using acoustic levitation (Figure 6A). AcoustoFab utilizes
shrinkage and selection operator, support vector regressor, OpenMPD (multimodal particle-based displays) and
95
extreme boosting gradient) on these features using 10- the boundary element method algorithms to enable
96
fold cross-validation (R = 0.977–0.978). The ensemble the formation of multiple acoustic traps in proximity
2
model achieved high predictive accuracy, demonstrating to complex sound-scattering surfaces for depositing
its potential to enhance the precision of droplet-based materials on these complex substrates. The OpenMPD
bioprinting, optimizing process parameters to improve algorithm dynamically calculates and updates the
reproducibility and scalability. positions of multiple acoustic tweezers, allowing control

AI-driven approaches trained on experimental over the levitation and omnidirectional movement of
datasets and computational simulations predict optimal multiple droplets of bioinks in mid-air. The boundary
printing conditions, ensuring high-resolution constructs element method algorithm simulates acoustic wave
with minimal defects and material waste. They reduce interactions with large, scattering surfaces in real-time,
the need for manual adjustments and trial-and-error predicting and mitigating potential distortions in the
experimentation, making bioprinting more efficient, acoustic field. This allows levitating, transporting, and in
scalable, and sustainable. While these developments are situ depositing of bioinks onto irregular surfaces of diverse
promising, current models, such as those by Bone et al., orientations, including a human hand. AcoustoFab is
85
87
Fu et al., and Chen et al., are often trained on material- capable of printing any soft materials within a wide range
86
specific or relatively small datasets, which may limit of viscosities (1–5,000,000 mPa·s), including biopolymers,
their immediate applicability across different bioprinting composite hydrogels, and bioinks. The embedded cells
platforms or bioink types. Some frameworks, like those in the hydrogels also demonstrated high viability post-
89
88
developed by Zhang et al. and Xu et al. have begun printing. The contactless nature of AcoustoFab minimizes
to incorporate cell viability and rheological behavior, cross-contamination, mechanical wear, and substrate
reflecting a growing interest in integrating both physical damage, contributing to reduced material waste and lower
and biological outcomes into predictive models. Others, failure rates, which is critical in clinical applications. By
like Rojek et al., demonstrate the potential of AI in integrating advanced algorithms with acoustic levitation,
90
reducing material waste and energy use, aligning with AcoustoFab presents a promising approach for intelligent
broader sustainability goals. Looking ahead, expanding and sustainable biofabrication. 97

Volume 11 Issue 4 (2025) 143 doi: 10.36922/IJB025170164