Page 315 - IJB-8-4

RESEARCH ARTICLE

A Deep Learning Quality Control Loop of the

Extrusion-based Bioprinting Process

Amedeo Franco Bonatti , Giovanni Vozzi , Chee Kai Chua , Carmelo De Maria *
1
2
1
1
1 Department of Information Engineering and Research Center “Enrico Piaggio,” University of Pisa, Pisa, Italy
2 Engineering Product Development Pillar, Singapore University of Technology and Design, Singapore
Abstract: Extrusion-based bioprinting (EBB) represents one of the most used deposition technologies in the field of bioprinting,
thanks to key advantages such as the easy-to-use hardware and the wide variety of materials that can be successfully printed.
In recent years, research efforts have been focused on implementing a quality control loop for EBB, which can reduce the
trial-and-error process necessary to optimize the printing parameters for a specific ink, standardize the results of a print across
multiple laboratories, and so accelerate the translation of extrusion bioprinted products to more impactful clinical applications.
Due to its capacity to acquire relevant features from a training dataset and generalize to unseen data, machine learning (ML) is
currently being studied in literature as a relevant enabling technology for quality control in EBB. In this context, we propose
a robust, deep learning-based control loop to automatically optimize the printing parameters and monitor the printing process
online. We collected a comprehensive dataset of EBB prints by recording the process with a high-resolution webcam. To
model multiple printing scenarios, each video represents a combination of multiple parameters, including printing set-up
(either mechanical or pneumatic extrusion), material color, layer height, and infill density. After pre-processing, the collected
dataset was used to thoroughly train and evaluate an ad hoc defined convolutional neural network by controlling over-fitting
and optimizing the prediction time of the network. Finally, the ML model was used in a control loop to: (i) monitor the printing
process and detect if a print with an error could be stopped before completion to save material and time and (ii) automatically
optimize the printing parameters by combining the ML model with a previously published mathematical model of the EBB
process. Altogether, we demonstrated for the first time how ML techniques can be used to automatize the EBB process, paving
the way for a total quality control loop of the printing process.
Keywords: Extrusion-based bioprinting; Quality control; Convolutional neuronal network; Automatic parameter optimization

*Correspondence to: Carmelo De Maria, Department of Information Engineering and Research Center “Enrico Piaggio,” University of Pisa,
Pisa, Italy; carmelo.demaria@unipi.it
Received: August 26, 2022; Accepted: September 6, 2022; Published Online: October 11, 2022
(This article belongs to the Special Issue: Related to 3D Printing Technology and Materials)

Citation: Bonatti AF, Vozzi G, Chua CK, et al., 2022, A Deep Learning Quality Control Loop of the Extrusion-Based Bioprinting Process. Int
J Bioprint, 8(4):620. http://doi.org/10.18063/ijb.v8i4.620

1. Introduction mechanical (i.e., piston actuated and screw assisted)
or pneumatic, which is attached to a three-dimensional
In recent years, the field of bioprinting has seen a strong (3D) positioning system. By applying a pressure on the
increase of interest as a promising solution to fabricate material (usually a hydrogel) inside a reservoir (usually a
tissues and organs for tissue engineering applications [1,2] . syringe) while moving it over a printing plate, a 3D shape
Among the technologies currently available, extrusion- can be obtained in a layer-by-layer fashion [4,5] .
based bioprinting (EBB) has been adopted as the most Significant efforts have been made to develop
popular approach thanks to its simple and affordable novel biomaterial inks that show enhanced biological
[6]
hardware, a wide array of processable materials, and the response and appropriate physical-chemical properties
ability to print clinically sized constructs . The typical for the target tissue, while also being processable through
[3]
EBB set-up consists of an extrusion tool-head, either EBB with good printability (i.e., a combination of several
© 2022 Author(s). This is an Open-Access article distributed under the terms of the Creative Commons Attribution License, permitting distribution and
reproduction in any medium, provided the original work is properly cited.
307