Page 409 - IJB-10-3

International Journal of Bioprinting In situ thermal monitoring in bioprinting

similarity coefficient (DSC), a well-established measure Since the extruder of our bioprinter operated under
50
commonly employed in image analysis and medical the fundamental assumptions of consistent object
imaging applications, between the binarized target image brightness, limited displacement, and full visibility in all
(VR or IR) and the relative binarized nominal shape image: video frames, it was possible to apply a similar approach
to reconstruct the geometry of each layer of the “step”
XYI model, through an integration of the segmented pixel
DSC =⋅2 (I) obtained frame-by-frame.
X + Y
3. Results
where X is the set of foreground pixels in the target 3.1 Comparison of segmentation performance
image and Y is the corresponding set of foreground pixels between VR and IR images
in the binarized nominal shape image, respectively. The In the following, results obtained on a seven-layer construct
DSC value ranges between 0 and 1, where a value closer printed in the before-mentioned conditions were used as a
to 0 indicates less spatial overlap between regions X and Y, reference to show the promising advantage of IR images. In
while a value closer to 1 indicates a higher degree of spatial Figure 7, it is possible to see the images of each of the seven
overlap. By employing this metric, we aimed to quantify layers captured with the two types of cameras. Each image
the degree of similarity between the two image sets. demonstrates the respective reconstructed geometry and
the calculated DSC value. It is possible to notice that IR
2.8. Proof-of-concept of on-line geometric images led to a better geometry reconstruction, confirmed
reconstruction capability not only by visual inspection, albeit with all the relevant
Since different materials exhibit different thermal hardware resolution limitations, but also by the DSC
properties, during the second campaign, it was also decided values obtained, which were always higher than those
to implement the developed segmentation algorithm along obtained for VR images. Furthermore, it is possible to
with a tracking algorithm, which relied on the Kanade– identify trends in the performance of the metrics in the
Lucas–Tomasi (KLT) algorithm, to perform segmentation two different sets of images.
of the deposited construct at the time of maximum thermal
gradient between background and foreground (Figure 6), 3.2. Last layer detection
to enhance the monitoring capabilities of our system. For In the course of the experimental campaigns, several
further details on algorithm implementation, we suggest samples quite frequently suffered from printing problems,
the reader refer to a previous work of ours, where the which resulted in defects on their appearance. Figure 8
51
KLT algorithm, renowned within the field of computer shows the original VR image, the original IR image, and
vision for its effectiveness as a feature-based tracking the IR segmented image of the third layer of a three-layer
method, showed excellent performance in the tracking of construct that, despite being printed with the same printing
the extruder of a 3D printer across consecutive thermal parameters as the other samples, ran into under-extrusion
video frames. problems, presenting a “pillar” geometry in the last printed

Figure 6. On the left, the original IR frame from a video of the second campaign. On the right frame-by-frame integration of the segmented pixel once
extruded from the nozzle of the first instants of the process.

Volume 10 Issue 3 (2024) 401 doi: 10.36922/ijb.2021