3.1 Automating the processing of CT scans by training a neural network
To train a neural network to automatically clean a set of CT-slices (DICOM file), we need to manually prepare cases for its training. By cases we mean sets of two CT slices: the original ones (resulting from CT scans, uncleaned) and the ones cleared of artifacts (cleaned). Unnecessary artifacts are usually existed in CT scans due to foreign objects such as implants or orthodontic anchor and various physical and chemical processes (Kokorev et al. 2008). Such scans require additional processing prior to bone modeling, which is very time-consuming. We chose U-net neural network for acceleration and automation of CT scan processing.
To clean the slices, we first create a 3D model of them using 3D Slicer and Blender, then select and remove artifacts in Zbrush. Then we use Photoshop to cut the 3D model back into slices. Now we have two sets of slices (cleaned and uncleaned), which we use to train the neural network. Each training epoch we create dataset from slices by making graphic augmentation on each slice. These datasets used for learning U-Net with optimizer adam and binary-cross entropy loss. After 12 epochs we have trained neural network to automatically clear CT scans from artifacts. The result of neural network operation is shown in the Figure 1.
In this way before using our trained network the CT scan processing took a significant amount of time. It took an average of 1 hour to process CT scan in 3D Slicer, together with 3-5 hours in Blender and Photoshop, together with 35 minutes in Zbrush. Total from 4.5 hours up to 6.5 hours. Using our trained network on average now it takes 1 minute to get a clean 3D jaw model.
After obtaining a cleaned 3D jaw model, we manually design the membrane. In ZBrush we create a cylinder, place it on the site of the defect (Fig. 2 - left image), use different brushes to shape it (Fig. 2 - middle image), relying on the rules of anatomy so that the edge of the membrane continuously touches the existing bone and create technical holes (Fig. 2 - right image), for screws and filling the bone grafting material.
maxillofacial surgery, traumatology, surgery (Pavlovskij et al. 2017; Kotelnikov et al. 2017), this method allows the most precise restoration of bone tissues, taking into account the individual anatomical features of each patient. CT scans are often contaminated with unnecessary artifacts (Kokorev et al. 2008). Such scans require additional processing (Kidoh et al. 2014) prior to bone modeling. This is time-consuming for medical professionals. Neural networks of different architectures are widely used for acceleration and automation of CT scan processing (Muge et al. 2018), U-net, which we use for CT scan processing, shows a high performance, taking into account the large volume of data to be processed (Šerifović Trbalić et al. 2019). Also, automation of CT scan processing helps to avoid human errors and human misinterpretation of the results (Xue et al. 2019).
3.2 Laser-induced formation of an antibacterial coating
Two mechanisms can be distinguished for the formation of oxide films on metal surfaces during laser heating by pulses of nanosecond duration. In the first case, when heating the metal surface below the boiling point, the oxide layer is formed at both the heating and cooling stages (Veiko et al. 2014). In the second case, when heated above the boiling point of the metal, at the stage of heating material evaporates, and at the stage of cooling an oxide film is formed. Depending on the duration of the cooling stage, it is possible to control its chemical composition and thickness (Veiko et al. 2014). Therefore, laser exposure modes from both temperature ranges were chosen to form oxide films with different chemical composition and thickness of oxide films. In our experiments heating the metal surface above the boiling point is realized by using high frequency (900 – 999 kHz), which corresponds to the samples named Hy, Hb, Hg, Hp. Heating the metal surface below the boiling point is realized by using low frequency (99 kHz) and high frequency, which corresponds to the samples named Ly, Lp, Lb. The processing modes is presented in Table 1, where q is power density, and teff is the effective exposure time which is defined as the product of the number of pulses that arrived at a point in the irradiated space, multiplied by the pulse duration, and multiplied by the number of passes.
An indirect parameter characterizing the different chemical composition or thickness of the oxide films is the color of the surface after laser exposure (Figure 3).
Table 1 Laser processing parameters with two frequency ranges: low frequency (99 kHz) – L-samples, and high frequency (900 – 999 kHz) – H-samples
Name
|
Color
|
q, MW/cm2
|
teff, μs
|
Hy
|
Yellow
|
10
|
14
|
Hb
|
Blue
|
10
|
29
|
Hg
|
Grey
|
10
|
77
|
Hp
|
Pink
|
10
|
165
|
Ly
|
Yellow
|
14
|
25
|
Lp
|
Purple
|
14
|
50
|
Lb
|
Blue
|
14
|
74
|
3.3 Research on Singlet Oxygen Generation on Oxide Coatings
Below are graphs of the luminescence spectra of the samples in the region of 1240-1300 nm. From the analysis of these spectra, we can see that the luminescence peak at a wavelength of 1270 nm is in the spectra belonging to the following samples: "Sample name-luminescence wavelength" - Hb-285, Hb-375, Hb-405, Hy-285, Hy-375, Hy-405, Hg-375, Hg-405, Hp-375, Ly-375, Ly-405, Lp-405, Lb-285, Lb-375 and Lb-405. Hb-285, Hy-285, Hy-375, Hg-375, Hp-375, Lb-285 and Lb-375 samples are characterized by a less pronounced peak at this wavelength. This means that the oxide films of Hb, Hy, Hg, Ly, Lp, and Lb have a high ability to generate singlet oxygen when exposed to excitation radiation at a wavelength of 405 nm, and the oxide films of Hb and Ly - at a wavelength of 375 nm.
Table 2 Graphs of the luminescence spectra of the samples in the region of 1240-1300 nm. Clear peaks are highlighted by circle
UV
|
285 nm
|
375 nm
|
405 nm
|
Hb
|
|
|
|
Hy
|
|
|
|
Hg
|
|
|
|
Hp
|
|
|
|
Ly
|
|
|
|
Lp
|
|
|
|
Lb
|
|
|
|
On samples of Hp and Hg, generation of singlet oxygen was not observed at almost all wavelengths (except Hg-405). In this regard, Hb, Hy, Ly, Lp and Lb were chosen for further studies of the physicochemical properties of the obtained oxide films.
3.4 Study of the morphology and composition of the samples
Scanning electron microscope (SEM) photographs (fig. 4) clearly show traces of laser radiation scanning on the surface of the samples. It can also be seen that under high-frequency modes, the oxide film is less prone to cracking, due to the shorter time of the surface oxidation.
An energy dispersive X-ray spectroscopy study was also carried out (the results are shown in text in the figure 4), which showed that the average percentage of oxygen in low-frequency modes is greater.
In the case of high frequency modes according to TEM images (fig. 5) the thickness of the oxide layer of the sample Hy about 17 nm, and Hb about 40 nm. Research method high-resolution electron microscopy showed that the oxide layer sample Hy has lattice parameters close in parameters to those of titanium oxide TiO2 (rutile), and the Hb sample consists of two layers: an inner layer of TiO2 (anatase) and outer layer Ti3O5, layer thickness 15 nm and 25 nm respectively.
In the case of low frequency modes according to the data of transmission electron microscopy, the thickness of the oxide layers for the Ly sample is about 15 nm, and for the Lp and Lb samples, about 25 nm. Studies by high-resolution electron microscopy have shown that the oxide layer of the Lp sample has lattice parameters close in parameters to titanium oxide TiO2 (rutile), and for the Ly and Lb samples, the oxide layer has lattice parameters close to those of titanium oxide TiO2 (anatase).
Experimental and literature data on the lattice parameters of TiO2 (rutile), TiO2 (anatase) и Ti3O5 are presented in Table 3.
Table 3 Experimental and literature data on the lattice parameters of TiO2 (rutile), TiO2 (anatase) and Ti3O5
hkl
|
TiO2 (anatase), Å
|
hkl
|
TiO2 (rutile), Å
|
hkl
|
Ti3O5, Å
|
experimental value
|
theoretical value [39]
|
experimental value
|
theoretical value [39]
|
experimental value
|
theoretical value [39]
|
Hb
|
Ly
|
Lb
|
Hy
|
Lp
|
Hb
|
101
|
3,54
|
3,49
|
3,6
|
3,52
|
110
|
3,26
|
3,3
|
3,25
|
201
|
4,32
|
4,29
|
103
|
2,41
|
2,44
|
2,3
|
2,43
|
101
|
-
|
-
|
-
|
110
|
3,57
|
3,54
|
004
|
2,37
|
2,39
|
-
|
2,37
|
200
|
2,31
|
2,32
|
2,3
|
-
|
-
|
-
|
202
|
1,78
|
-
|
-
|
1,75
|
-
|
-
|
-
|
-
|
-
|
-
|
-
|
213
|
1,49
|
-
|
-
|
1,49
|
-
|
-
|
-
|
-
|
-
|
-
|
-
|
The chemical composition and thickness of the oxide film of the samples are shown in Table 4.
Table 4 Resulting chemical composition and thickness of the oxide layer of the samples
Sample
|
Chemical composition
|
Oxide film thickness, nm
|
Hb
|
TiO2 anatase
|
7
|
Ti3O5
|
18
|
Hy
|
TiO2 rutile
|
12
|
Ly
|
TiO2 anatase
|
15
|
Lp
|
TiO2 rutile
|
25
|
Lb
|
TiO2 anatase
|
25
|
The composition of almost all samples is titanium dioxide in the polycrystalline modification of rutile and anatase. Titanium dioxide is a semiconductor, and its crystal morphotypes anatase and rutile exhibit high photoactivity when irradiated with ultraviolet light. When a photon with energy hν is absorbed, electrons move from the valence zone to the conduction zone. There is an electron transfer to the oxygen molecule with the formation of a superoxide ion radical. Superoxide ion formation is a basic process for UV-activated titanium oxide. However, there are various further processes with the formation of other radical ions on the surface, which should also be considered in connection with their antibacterial properties.
In our previous article (Doll et.al. 2021) the Hb sample (compared to the Hy sample) showed the best antibacterial activity after irradiation at a wavelength of 375 nm. Indeed, as can be seen from Table 2, the intensity of the peak of singlet oxygen at a wavelength of 375 nm for the sample Hb is much stronger than for the sample Hy. But, if we choose a wavelength of 405 nm for irradiating the membrane, then the peak intensity for all samples will be almost identical, which should positively affect the antibacterial properties of the membrane and make it possible to paint it in different colors for decorative purposes. At this wavelength, the polycrystalline modification of titanium dioxide does not play a role.
3.5 Laser-induced formation of an antibacterial coating on the membrane surface
The dental membrane is U-shaped. We divided its surface into three regions - two lateral and one upper. For laser processing of the membrane surface, we equipped the laser complex by single-axis rotator, which performed a 90˚ rotation of the fixed membrane, as shown in Figure 6a. When processing each of the three surfaces of the membrane, the laser beam was moved along a rectilinear trajectory, performing line-by-line scanning of a rectangular area in the focusing plane. The Hb laser treatment mode was used. The result of laser processing is shown in Figure 6b.