The effect of cryopreservation on enamel microcracks – A μCT analysis using a deep learning algorithm

Abstract

To date, the effect of cryopreservation on microcracks in the dental enamel remains unclear. These enamel microcracks are very thin, at the limit of visibility and their segmentation is beyond the capabilities of traditional image analysis. The objective of the present study was to investigate the effect of cryopreservation on enamel microcracks with a μCT analysis using a deep learning algorithm.

A manual annotation was performed to construct a high-quality training and testing dataset. A 4-phase semantic segmentation U-Net architecture neural network was trained in Dragonfly (ORS systems, Canada) and then applied on a sample of 5 teeth before and after cryopreservation, enabling for the first time a direct evaluation of the formation and evolution of enamel cracks caused by cryopreservation.

Qualitatively, the segmentation results were very satisfying and cracks as thin as 2–3 voxels wide could be segmented automatically. All teeth presented enamel microcracks, without propagation into the dentin. Similar crack patterns were observed in all teeth and may be related to the use of forceps during extraction. In the post-cryopreservation scans the damage extended, and new smaller cracks appeared on the occlusal surface of the tooth. Quantitatively, the average crack/enamel ratio was 0.066 ± 0.021 % before cryopreservation, and 0.087 ± 0.018 % after cryopreservation.

The present study presents the first scalable yet precise method to quantify clinically relevant tooth damage after cryopreservation. As such, this method also opens the way to future in-depth studies on dental enamel in many dental fields.