Publikationsdatenbank

Artificial intelligence versus manual quantification of angiogenesis (2022)

Art

Poster

Autoren

Alshamy, Zaher (WE 1)

Plendl, Johanna (WE 1)

Kässmeyer, Sabine

Kongress

33rd virtual conference of the European Association of Veterinary Anatomists

Ghent, Belgium, 28. – 31.07.2021

Quelle

Anatomia, histologia, embryologia

Bandzählung: 51

Heftzählung: S1

Seiten: 5

ISSN: 0340-2096

Sprache

Englisch

Verweise

URL (Volltext): https://onlinelibrary.wiley.com/doi/10.1111/ahe.12759

DOI: 10.1111/ahe.12759

Kontakt

Institut für Veterinär-Anatomie

Koserstr. 20
14195 Berlin
+49 30 838 75784
anatomie@vetmed.fu-berlin.de

Abstract / Zusammenfassung

Introduction:
Angiogenesis is a physiological process through which new blood vessels are generated from pre-existing vasculature. Morphological parameters characterizing vascular networks in vitro can be evaluated by different methods. The aim of this study was to compare the conventional manual with a new Artificial Intelligence (AI) based method for quantification of angiogenesis.

Materials and Methods:
An AI module (Segment.ai by Nikon, Düsseldorf, Germany) has been trained on a small set of hand-traced microscopic images. The training outcome was applied on similar im-ages, to automatically recognize structures previously only identifia-ble by manual tracing. Human dermal endothelial cells were cultured and labelled with the endothelial marker anti-CD31. Cells prolifer-ated and formed a 3D tubular network. Number and diameter of “endothelial tubes” and the points of their crossing (“knots”) were quantified manually as well as by Segment.ai.

Results:
Quantification of number and diameter of endothelial tubes yielded similar results with both methods. Mean of number and diam-eter of endothelial tubes was (835.17 ± 52.37 SEM, 9.91 μm ± 0.21 SEM, respectively) for the manual method and (865 ± 103.58 SEM, 10.02 μm ± 0.80 SEM, respectively) for AI. Number of knots was 323 ± 26 for manual and 610 ± 159 for AI. Training of AI took about 50 hours; time for quantification of each image was roughly 40 min-utes for manual and 5 minutes for AI.

Conclusion:
AI saves time and effort – provided the user is well ac-quainted with the method. Training is time consuming and results are still impaired by artefacts.