To date, the available information on teeth MCs and their characteristics has been obtained using 2D analysis techniques such as stereomicroscopy7,8,9,10, scanning electron microscopy (SEM)1, 3, 11 or three-dimensional (3D) scanning methods (optical coherence tomography (OCT) and ultrasound)12,13,14,15,16. The evaluation and measurement of qualitative and quantitative parameters of cracks (their number, direction, location, length, and width) presented in previously published literature describe the morphology and behaviour of MCs only on the outer enamel surface1, 3, 7,8,9, 17,18,19. Thus, it is still unknown whether these cracks are limited to the enamel or whether they can extend beyond the DEJ into the dentin or even the pulp.
Cube Iq 4 0 Full Crack 71l
Several studies have been carried out so far on the depth parameter of cracks13, 15, 20,21,22. However, the limitations of the techniques used in those studies (e.g. the limited depth of penetration and scanning range of the device utilized for MCs analysis15, 20, sensitivity of the technique to surface curvature13, 20, the depth measurements carried out on a simulated human tooth13, an indirect method of determining the depth of the crack21, 22, the need for crack infiltration with contrast material for depth assessment22, or the physical measurement of the crack after cutting the tooth21) have all been the reasons behind the search and development of a new approach enclosing 3D imaging technique that would enable a non-destructive examination of MCs with micrometer resolution.
A more detailed description of the scanning procedure is available in a previously published study34. Here we provide a brief summary of experimental set-up. X-ray \(\mu\)CT scans with X-ray microscope (Xradia 520 Versa; ZEISS, Pleasanton, CA, USA) were used in the current study of a 3D distribution of cracks and other features in the teeth samples. The general layout of the experiment consisted of the X-ray source, the specimen, and the detector. The distances between the source-sample and sample-detector were adjusted to achieve the maximum magnification with the full field-of-view of the tooth sample. For this experiment the following distances were used: source-sample = \(21\,\hbox mm\) and sample-detector = \(125\,\hbox mm\). The sample projection images were obtained in absorption mode using geometrical magnification and a CCD detector resulting in a detector size of 2048 \(\times\) 2048 pixels. To achieve the optimal signal-to-noise level (intensities of >5,000 grey value over low transmission regions), the exposure time of \(5\,\hbox s\) was selected. The distances between the source, sample and detector resulted in a \(5\,\times 5\,\times 5\, \upmu \hbox m^3\) voxel size which in this case defines the experimental resolution34. The final result of the scanning procedure was four data-cubes of \(\sim\)10\(^3\) mm\(^3\) (\(\sim\) \(2000^3\)) voxels, containing values stored as 16-bit integers, with voxel edge of \(\sim\) \(5\,\upmu \hbox m\).
For faster visualization purposes, the data-cubes were resampled to a common voxel scale and aligned to principal contact surfaces of the teeth with \(\sim\)10 \(\upmu \hbox m\) scale. In Fig. 2 axes x, y, z were aligned as indicated (x along the palatal, y along the contact, and z along the vertical extent of tooth). For detailed mapping of MCs, the vertical slice of the tooth was divided into three horizontal slabs of equal height corresponding to 1/3 (cervical third), 2/3 (middle third) and 3/3 (occlusal third) of the tooth surface. The division of the tooth surface was based on dental examination methodologies and different enamel quality and mechanical properties of the individual tooth (Fig. 2)1, 4, 7, 36, 39,40,41.
Although the principal components of the tooth can be identified with a naked eye by the grey level of voxel value, such a straightforward selection of voxels to isolate tooth components would result in a rather poor quality, e.g. cracks, pulp, and outside of the tooth would have similar numerical values. Therefore, we trained a CNN to identify voxels (pixels in each slice since we processed data-cube as slices for this purpose) which belong to these four categories: (1) cracks, (2) enamel, (3) dentin, (4) air.
As an input for training, slices of the tooth along z axis of 512 \(\times\) 512 pixels were cropped and converted to RGB image repeating a 2D greyscale array three times, making it suitable for the ResNet50 input. As target labels, an array with the same spatial size, but with four channels (one-hot encoding the categories) was created, with any given pixel attributed to either cracks, enamel, dentin, or air. Categorical cross entropy loss function was used during training.
Four healthy undamaged human maxillary premolars (with and without visible enamel MCs on the outer tooth surface) that had been extracted for orthodontic reasons were analyzed using \(\mu\)CT together with CNN assisted segmentation. X-ray images of all the samples showed a dense tooth structure in which enamel, dentin, pulp, and cracks could be identified. The teeth appeared to be cracked, but without visible damage or separation of fragments. The network of cracks found in all the healthy teeth examined suggests that the cracks, along with the enamel, dentin, and pulp, could be considered a structural element of the tooth. The summary of the study is shown graphically in Fig. 4 with detailed results presented below.
Example of a healthy human tooth extracted for orthodontic reasons (a). (b) X-ray \(\mu\)CT data-cube projected density map of the tooth with visible enamel, dentin, pulp, and cracks. (c) Voxel density projection of convolutional neural network segmented cracks, enamel, and dentin (red, green, blue) revealing intricate inner structure and suggesting that cracks could be considered an integral part of the tooth along with enamel, dentin, and pulp.
The applied scanning technique and developed segmentation approach allowed us to analyze the arrangement of MCs in all four teeth samples. Cracks that connect to each other have been identified and differentiated from those that are isolated. A single network of star-shaped cracks (longitudinally in relation to the main axis of the tooth) was found to cover most of the internal tooth structure (Fig. 7, see Supplementary Movie 3). This continuous connective formation occupies \(\sim\)2% of the volume of a tooth. In contrast, the remaining groups of single unconnected cracks tend to be located closer to the outer surface of the tooth and occupy a significantly smaller volume.
The 3D visualization of the four teeth allowed us to evaluate the structural properties of MCs in each sample (Fig. 8, Fig. 9, and Supplementary Movie 4). In the microcrack network, it was possible to distinguish the main planes of the crack in two almost perpendicular directions, thus revealing the crack as a planar (interconnected manifolds) rather than a threaded structure (Fig. 7).
The proposed method is insensitive to the curved surfaces of the specimens used in the study, and as a result, it overcomes the shortcomings of the previous techniques of crack analysis13, 20. This is particularly important when the subject is a tooth having four surfaces of different convexity, of which the buccal surface is the most commonly examined and also the most convex. Compared to the 3D scanning methods used so far (OCT, ultrasound)15, 20, X-ray \(\mu\)CT allows the assessment of MCs at various distances from the outer enamel surface (enamel thickness \(\approx\) \(0.5\,\hbox mm\) in the cervical region, up to \(\approx\) \(2.5\,\hbox mm\) near the cusp for the molar teeth)40 or even in deeper layers of the tooth, e.g. the dentin, and is not affected by the different densities of these materials (density of enamel, \(2.61\pm 0.04\, \hbox g/cm^3 -- 2.77\pm 0.04\, \hbox g/cm^3\), and dentin \(1.79\pm 0.02\, \hbox g/cm^3 -- 2.12\pm 0.03\, \hbox g/cm^3\))53.
MCs can be confused with developmental defects of enamel (e.g. enamel lamellae). However, studies have shown that cracks and enamel lamellae are not identical68. Enamel lamellae can be described as fluid-filled cracks in the enamel (containing organic substances) that extend from the DEJ to the surface of the enamel, or vice versa68, 69. It has been observed that during the decalcification, the cracks disappear and the lamellae persist as a coherent organic, sheetlike process68. Meanwhile, the enamel cracks of the extracted tooth remain empty as no organic material is left68, 69. It should be noted that no new studies on lamellae have been carried out in the last decade. Therefore, in order to update the available information on enamel lamellae and to further elucidate the differences between cracks and lamellae, it is necessary to analyze enamel developmental defects using state-of-the-art techniques.
The available literature presents evidence of the correlation between the tissue dehydration and the dynamic dimensional changes within dentin and enamel70, as well as between dehydration and the fatigue crack growth resistance71. Due to the characteristics of the X-ray scanner used in this study34, the samples could not be stored in an aqueous media during the scanning procedure. Although it is not known exactly what effect the storage of the samples in non-hydrated media may have had on the cracks located in the crown of a tooth, it has already been shown that the width and length values of enamel MCs are not affected by the dehydration that occurs during the preparation of the samples for SEM scanning and observation (no new MCs were registered either)1, 72.
Our study had limitations. Firstly, the small sample size prevented us from drawing more general conclusions. However, by combining X-ray \(\mu\)CT together with machine learning, we have presented a novel method to analyze the 3D tooth microstructure and verified it using four extracted human teeth. Thus, we believe that despite the small number of specimens, the demonstrative value of our experimentally validated approach is high. In the future, we plan to test the method on a larger sample size, which would also increase the probative value of the work and its reliability as well as its versatility. Secondly, even after an atraumatic tooth extraction procedure, it cannot be guaranteed that cracks will not develop. It has to be acknowledged that the experience of an oral surgeon reduces, but does not eliminate, the possibility of new MCs. However, every effort was made to select only healthy, undamaged teeth as our study sample. Thirdly, due to the technical characteristics of the X-ray scanner, the specimens could not be stored in an aqueous media during the scanning procedure. Although it is not known exactly to what extent dehydration can affect teeth cracks, the inability to avoid it completely during the study could be considered as one of the limitations of our work. Finally, training the segmentation model is a subjective task, requiring visual decisions when selecting/rejecting groups of voxels and attributing labels to them, as there is no exact ground-truth, and the whole process is limited by scanning and reconstruction technique, noise, and error propagation, affecting classification power of the model, which could be increased once volume segmentation is performed with 3D convolution kernels instead of slice-by-slice 2D aggregate. 2ff7e9595c
Comments