Objective. To assess the influence of light emitting diode (LED) and quartz tungsten halogen (QTH) light curing unit (LCU) on the bottom/top (B/T) Vickers Hardness Number (VHN) ratio of different composites with different shades and determination of the most significant effect on B/T VHN ratio of composites by shade, light curing unit, and composite parameters using artificial neural network. Method. Three composite resin materials [Clearfil Majesty Esthetic (CME), Tetric N Ceram (TNC), and Tetric Evo Ceram (TEC)] in different shades (HO, A2, B2, Bleach L, Bleach M) were used. The composites were polymerized with three different LED LCUs (Elipar S10, Bluephase 20i, Valo) and halogen LCU (Hilux). Vickers hardness measurements were made at a load of 100 g for 10 sec on the top and bottom surfaces and B/T VHN ratio calculated. The data were statistically analyzed with three-way ANOVA and Tukey test at a significance level of 0.05. The obtained measurements and data were then fed to a neural network to establish the correlation between the inputs and outputs. Results. There were no significant differences between the B/T VHN ratio of LCUs for the HO and B shades of CME (p>0.05), but there were significant differences between the B/T VHN ratio of LCUs for shade A2 (p<0.05). No significant difference was determined between the B/T VHN ratio of LCUs for all shades of TNC (p>0.05). For TEC, there was no significant difference between the B/T VHN ratio of halogen and LED LCUs (p>0.05), but a significant difference was determined among the LED LCUs (p<0.05). The artificial neural network results showed that a combination of the curing light and composite parameter had the most significant effect on the B/T VHN ratio of composites. Shade has the lowest effect on the B/T VHN ratio of composites. Conclusion. The B/T VHN ratio values of different resin-based composite materials may vary depending on the light curing device. In addition, the artificial neural network results showed that the LCU and composite parameter had the most significant effect on the B/T VHN ratio of the composites. Shade has the lowest effect on the B/T VHN ratio of composites.
The polymerization of resin-based composites is still generally based on light activation of camphorquinone (CQ) [
Lucirin TPO is now used in some composites because it is completely colorless after the light curing reaction, and its polymers are less yellow than others in which only camphorquinone is used as a photoinitiator. When a bleached tooth needs to be restored, the reduction of discoloration related to the photoinitiator is clinically significant in order to obtain and maintain color in aesthetic restorations [
The degree of conversion (DC) of dental resin composites is crucial in determining the physical/mechanical performance of the material and its biocompatibility. Strength, modulus, hardness, and solubility are directly related to the DC [
Artificial neural networks have numerous applications in scientific and social applications and their predictive abilities can cause many problems. They possess great opportunities for problems without clear mathematical linkages between the inputs and outputs. While neural networks can successfully predict many problems, additional caution must be taken to develop meaningful neural network structures, since by nature, neural networks do not recognize the physical meaning of inputs and outputs. Inadvertently designed neural network structures may not have the required generalization ability when showing applicable results for their original training data [
The aim of this study is to assess the influence of three different LED LCUs and a conventional QTH LCU on the bottom/top (B/T) Vickers hardness (VH) ratio of different composites with different shades. Another aim is to determine which parameters among shade, type of composite and type of light cure device has the strongest effect on the B/T VHN ratio of composites using an artificial neural network. The null hypothesis tested was that different LCUs did not affect the B/T VHN ratio of different composites with different shades.
In the present study, three different composite resin materials [Clearfil Majesty Esthetic (CME) (Kuraray, Osaka, Japan), Tetric N Ceram (TNC) (Ivoclar Vivadent AG, Schaan, Liechtenstein), and Tetric Evo Ceram (TEC) (Ivoclar Vivadent AG, Schaan, Liechtenstein)] with different shades were used (Table 1). For each tested material, 20 cylindrical specimens (2-mm-depth and 5-mm-diameter) were prepared using metallic molds. In order to obtain a flat polymerized surface, the specimens were covered on both sides with a polyester matrix strip and a thin, rigid microscope slide and photopolymerized with a conventional QTH (900-1100mW/cm
Materials and their composition.
Material Type Organic Matrix Inorganic Matrix Photoinitiator Shades Clearfil Majesty Bis-GMA, Barium glass, silica Camphorquinone HO Esthetic TEGDMA (85.5 wt%) (468 nm) A2 B2 Tetric N Ceram Bis-GMA- Barium glass, Lucirin TPO3 Bleach L (Ivoclar UDMA (15%), ytterbium trifluoride, (350-425 nm) Bleach M Vivadent AG, Bis-EMA (3.8%) oxides, silicon + Schaan, dioxide (63.5%) Ivocerin Liechtenstein) prepolymers (17%) (370-460 nm) + 81wt %, 55-57 vol% camphorquinone (468 nm) Tetric Evo Bis-GMA, Barium glass, Lucirin TPO3 Bleach L Ceram (Ivoclar UDMA, ytterbium trifluoride, (350-425 nm) Bleach M Vivadent AG, Bis-EMA mixed oxide (48.5%) + Schaan, (16.8%) prepolymers (34%) Ivocerin Liechtenstein) (370-460 nm) 80 wt%, 61 vol% + camphorquinone (468 nm)
In this study, different neural networks were employed using Matlab and their performances were evaluated to find the most successful network structure. Three different inputs and six outputs were used in this study, as shown in Table 2. Since the inputs of the study were nonnumeric, the enumeration technique was employed to use the inputs with the numeric output values by giving a unique number to every component used in the study. All of the inputs and outputs were normalized in the neural networks' operations. Different numbers of hidden layer neurons and different training functions, namely, Levenberg-Marquardt, Bayesian regulation, scaled conjugate gradient, and resilient backpropagation, were employed to find the network structure with the best predictive performance. Tangent sigmoid transfer functions were used for both layers. The Nguyen-Widrow initialization function was used for weights and biases to minimize the computation time. In order to prevent overfitting, appropriate convergence criteria were selected, and the validation process was monitored during the training phase.
Inputs and outputs of the network.
Input 1 Input 2 Input 3 Outputs Composite Shade Curing Unit - Upper 1 - Clearfil Majesty esthetic - B2 Hilux - Upper 2 - Tetric N ceram bleach - HO Bluephase 20i - Upper 3 - Tetric Evo ceram bleach - A2 Valo - Lower 1 - M Elipar S10 - Lower 2 - L - Lower 3
The performance of the tested networks was determined by using mean square error (MSE) and correlation coefficient (R) values, which were defined as follows: (
Three-way ANOVA results showed that there were significances for composites (p=0.001), LCUs (p=0.001), and composite∗LCU interaction (p=0.001). No significance was observed for shade (p=0.328), composite∗shade interaction (p=0.807), shade∗LCU interaction (p=0.364), and composite∗shade∗LCU interaction (0.531). (Table 3)
Three-way ANOVA results to compare the dependent variable B/T VHN ratios for the fixed factors of three different composite resins, four different LCUs and five different shades and their interactions at a significance level of 0.05.
Dependent Variable: Bottom/top ratio Source Type III Sum of Squares df Mean Square F Sig. Corrected Model 28014,147 a 27 1037,561 7,354 ,000 Intercept 1213334,731 1 1213334,731 8600,098 ,000 Composite 3076,818 1 3076,818 21,808 ,000 Shade 491,862 3 163,954 1,162 ,328 LCU 9541,274 3 3180,425 22,543 ,000 Composite ∗ Shade 8,443 1 8,443 ,060 ,807 Composite ∗ LCU 5660,444 3 1886,815 13,374 ,000 Shade ∗ LCU 1404,585 9 156,065 1,106 ,364 Composite ∗ Shade ∗ LCU 312,873 3 104,291 ,739 ,531 Error 15801,388 112 141,084 Total 1257150,267 140 Corrected Total 43815,536 139
Tukey post hoc test showed that there are significant differences between all composite materials (p<0.05). There was no significant difference between QTH LCU and Bluphase 20i (p=0.576) but there were significant differences between all other LCUs (p<0.05). When the shades were compared, there was no significant difference between B2 and HO, B2 and A2, A2 and HO, and M and L shades (p>0.05) but there were significant differences between all other shades (p<0.05). Multiple comparisons for composite, shade and LCUs are given in Tables 4–6, respectively.
Multiple comparisons of composite resin materials.
Multiple Comparisons Dependent Variable: Bottom/top ratio Tukey HSD (I) Composite (J) Composite Mean Difference (I-J) Std. Error Sig. 95% Confidence Interval Lower Bound Upper Bound Clearfil Tetric N Ceram -8,3398 ∗ 2,42456 ,002 -14,0988 -2,5809 Tetric Evo Ceram -20,7431 ∗ 2,42456 ,000 -26,5021 -14,9841 Tetric N Ceram Clearfil 8,3398 ∗ 2,42456 ,002 2,5809 14,0988 Tetric Evo Ceram -12,4033 ∗ 2,65597 ,000 -18,7119 -6,0946 Tetric Evo Ceram Clearfil 20,7431 ∗ 2,42456 ,000 14,9841 26,5021 Tetric N Ceram 12,4033 ∗ 2,65597 ,000 6,0946 18,7119
Multiple comparisons of shades.
Multiple Comparisons Dependent Variable: Bottom/top ratio Tukey HSD (I) Shade (J) Shade Mean Difference (I-J) Std. Error Sig. 95% Confidence Interval Lower Bound Upper Bound B2 HO 3,4773 3,75611 ,886 -6,9373 13,8920 A2 6,5826 3,75611 ,407 -3,8321 16,9972 M -10,3360 ∗ 3,25289 ,016 -19,3553 -1,3166 L -12,0404 ∗ 3,25289 ,003 -21,0597 -3,0210 HO B2 -3,4773 3,75611 ,886 -13,8920 6,9373 A2 3,1053 3,75611 ,922 -7,3094 13,5199 M -13,8133 ∗ 3,25289 ,000 -22,8326 -4,7939 L -15,5177 ∗ 3,25289 ,000 -24,5370 -6,4983 A2 B2 -6,5826 3,75611 ,407 -16,9972 3,8321 HO -3,1053 3,75611 ,922 -13,5199 7,3094 M -16,9186 ∗ 3,25289 ,000 -25,9379 -7,8992 L -18,6230 ∗ 3,25289 ,000 -27,6423 -9,6036 M B2 10,3360 ∗ 3,25289 ,016 1,3166 19,3553 HO 13,8133 ∗ 3,25289 ,000 4,7939 22,8326 A2 16,9186 ∗ 3,25289 ,000 7,8992 25,9379 L -1,7044 2,65597 ,968 -9,0687 5,6599 L B2 12,0404 ∗ 3,25289 ,003 3,0210 21,0597 HO 15,5177 ∗ 3,25289 ,000 6,4983 24,5370 A2 18,6230 ∗ 3,25289 ,000 9,6036 27,6423 M 1,7044 2,65597 ,968 -5,6599 9,0687
Multiple comparisons of LCUs.
Multiple Comparisons Dependent Variable: Bottom-top ratio Tukey HSD (I) LCU (J) LCU Mean Difference (I-J) Std. Error Sig. 95% Confidence Interval Lower Bound Upper Bound Halogen Bluephase -3,6400 2,83935 ,576 -11,0451 3,7651 Valo -11,6836 ∗ 2,83935 ,000 -19,0887 -4,2785 Elipar -21,4835 ∗ 2,83935 ,000 -28,8887 -14,0784 Bluephase Halogen 3,6400 2,83935 ,576 -3,7651 11,0451 Valo -8,0437 ∗ 2,83935 ,028 -15,4488 -,6386 Elipar -17,8436 ∗ 2,83935 ,000 -25,2487 -10,4385 Valo Halogen 11,6836 ∗ 2,83935 ,000 4,2785 19,0887 Bluephase 8,0437 ∗ 2,83935 ,028 ,6386 15,4488 Elipar -9,7999 ∗ 2,83935 ,004 -17,2050 -2,3948 Elipar Halogen 21,4835 ∗ 2,83935 ,000 14,0784 28,8887 Bluephase 17,8436 ∗ 2,83935 ,000 10,4385 25,2487 Valo 9,7999 ∗ 2,83935 ,004 2,3948 17,2050
When we split our data according to the B/T VHN ratios of the tested composite resin materials, the results are shown in Tables 7–9. Table 7 shows B/T VHN ratios of Clearfil Majesty Esthetic (CME). There were no significant differences between the LCUs for the HO and B2 shades of CME (p>0.05), but there was a significant difference between the LCUs for shade A2 (p<0.05). Hilux gave the lowest B/T VHN ratio for A2, and Elipar showed the highest B/T VHN ratio. No significant difference was determined between shade HO, B, and A2. Table 8 shows B/T VHN ratios for Tetric N Ceram Bleach (TNC). No significant difference was determined between the LCUs for all shades of TNC Bleach (p>0.05). No significant difference was determined between the Shade M and Shade L (p>0.05). Table 9 shows the B/T VHN ratios for Tetric Evo Ceram (TEC). There were no significant differences between the halogen and LED LCUs for both shades of TEC (p>0.05). When the LED LCUs were compared, Bluephase 20i LCU showed significantly higher B/T VHN ratio values than Elipar did (p<0.05). There was no significant difference between the Bluephase 20i and Valo groups (p>0.05). When the shades were compared no significant difference was determined between Shade M and Shade L.
B/T VHN ratio for Clearfil Majesty Esthetic.
Hilux Bluephase 20i Valo Elipar S 10 B2 86.7 Aa 82.7 Aa 85.1 Aa 97.9 Aa A2 68.6 Aa 76.3 ABa 88.7 Ba 92.5 Ba HO 77.2 Aa 76.7 Aa 90.7 Aa 93.9 Aa
B/T VHN ratio for Tetric N Ceram Bleach.
Hilux Bluephase 20i Valo Elipar S 10 Shade M 90.3 Aa 95.8 Aa 91.6 Aa 90.1 Aa Shade L 91.9 Aa 91.3 Aa 96.4 Aa 97.5 Aa
B/T VHN ratio for Tetric Evo Ceram Bleach.
Hilux Bluephase 20i Valo Elipar S10 Shade M 87.4 ABa 96.1 Aa 89.8 ABa 80.8 Ba Shade L 84.7 AB a 93.8 Aa 96.7 Aa 73.4 Ba
The best network structure to fit the data had 10 neurons in its hidden layer. Bayesian regulation was employed as the training function for the network. The MSE value of the network was 0.0373. The R values for the upper 3 and lower 3 outputs were 0.7487, 0.7890, 0.7721, 0.7674, 0.7674, and 0.7674, respectively. All figures, in subsequent chapters, as shown in Figures 1, 2, and 3 and Figures 4, 5, and 6, were drawn by using this neural network.
PHOTO (COLOR): Predictability of ANN model for upper side according to measurement 1.
PHOTO (COLOR): Predictability of ANN model for upper side according to measurement 2.
PHOTO (COLOR): Predictability of ANN model for upper side according to measurement 3.
PHOTO (COLOR): Predictability of ANN model for lower side according to measurement 1.
PHOTO (COLOR): Predictability of ANN model for lower side according to measurement 2.
PHOTO (COLOR): Predictability of ANN model for lower side according to measurement 3.
Dependency analysis was conducted to find the most influential parameter over the problem's outputs. The problems' inputs were fed to the neural network independently, and their performance was observed to reveal their impact on the outputs. The obtained results can be seen in Table 10. When single inputs were compared, curing light had the most significant effect, because it had the highest R values than the other two inputs. Shade has a lowest effect on the B/T VHN ratio of composites. A combination of composite and curing light inputs resulted in the highest R value among all of the combinations and had the most significant effect.
Dependency analysis of inputs.
Neural network inputs R Composite Shade Curing Unit Upper 1 Upper 2 Upper 3 Lower 1 Lower 2 Lower 3 X X X 0.7487 0.7890 0.7721 0.7674 0.7674 0.7674 X 0.2207 0.2191 0.2497 0.2654 0.2655 0.2654 X 0.1299 0.0680 0.1493 0.1795 0.1795 0.1795 X 0.4459 0.4808 0.5126 0.3927 0.3927 0.3928 X X 0.2080 0.1702 0.2501 0.2452 0.2452 0.2452 X X 0.6222 0.6549 0.6642 0.6494 0.6494 0.6494 X X 0.1039 0.0875 0.1538 0.2444 0.2444 0.2443
In the present study, the B/T VHN ratio of different composites were determined when using different shades and when polymerized with different LCUs. The results showed that the B/T VHN ratios of the different composite materials with different shades varied depending on the light curing device used. Thus, the null hypothesis was rejected.
In the present study, LCUs with different light intensities and wavelengths were used to polymerize composites. The results show that the HO and B2 shades of the CME composite specimens showed similar B/T VHN ratios when polymerized with different LCUs. However, A2 shades of CME showed higher B/T VHN ratio when polymerized with Valo and Elipar than when polymerized with Hilux. This may be because of the different composition and photoinitiator ratio of CME in different shades. However, exact photoinitiator ratios and composition of these materials in different shades were not obtained from the manufacturer.
The lower transmittance of the light results in a low DC and consequently low microhardness, which is strongly influenced by the resin's opacity and its filler contents. However, different shades of all composites used in the present study showed similar B/T VHN ratios when polymerized with the same LCU.
Alternative photoinitiators like Lucirin TPO and Ivocerin have recently been added to composite resins [
It is well proven that neural networks are suitable for complex problems that require extensive mathematical modelling. They can also be used on nonnumerical data which allows vast application areas such as advertisement, medicine, and sociology to use the extensive computation capability for generalization and prediction purposes. However special care must be taken when analyzing neural networks since network results would be accurate only as the size of the training data set. Therefore, data set must be selected such that the entire variable range is well presented, and necessary amount of data is present. As a result, this necessity requires extensive amount of data or experiments for a proper neural network training. Although satisfactory results and good generalization capability can be obtained, neural network should not be considered as a mathematical model by any means. This limits the use of neural network since the nature of the problem can not be solely determined from the network. It is also well known that the neural networks tend to give unsatisfactory results on data outside of the training data set or other data showing extreme characteristics. Acceptable results can be acquired for most of the complicated problems with carefully designed and trained networks however [
For sufficient polymerization, three vital characteristics are essential for a LCU: adequate light output, adequate wavelength range of light, and efficient exposure time [
In a microhardness study, Sabatini [
The efficiency of light curing techniques has often been assessed by depending on hardness measurements on the top and bottom surfaces of light-cured resin composite samples, and a bottom-to-top hardness ratio of 0.8 has been generally used as a standard for sufficient degree of cure [
Based on the current findings and within the limitations of this in vitro study, the B/T VHN ratios of different resin-based composite materials may vary depending on the light curing device. In addition, the artificial neural network results showed that the LCU and composite parameter had the most significant effect on the B/T VHN ratio of the composites. Shade has the lowest effect on the B/T VHN ratio of composites.
The data used to support the findings of this study are available from the corresponding author upon request.
This research received no specific grant from any funding agency in the public, commercial, or not for profit sectors.
The authors declare that there are no conflicts of interest regarding the publication of this paper.
By Hacer Deniz Arısu; Evrim Eligüzeloglu Dalkilic; Fehime Alkan; Sebnem Erol; Mine Betul Uctasli and Alican Cebi