In vitro validation of Digital Image Analysis Sequence (DIAS) for the assessment of the marginal fit of cement-retained implant-supported experimental crowns

In fixed prosthodontics, it is critical that marginal fit (MF) is kept to a minimum, for successful long-term survival of the prosthesis [1‐3]. Poor MF on the contrary, might accelerate cement dissolution and plaque retention, therefore increasing the risks of biological complications [4‐7]. Additionally, it drastically increases the probability of luting material exposure at the restoration margin [8, 9].

Clinically accepted MF ranges from as low as 26 μm and might exceed 200 μm [10‐16]. Although discrepancies at the margin, ideally, are ought to be kept to a minimum, perfect margins and absolute passive framework fit are unlikely to be achieved [17, 18].

Cementation is associated with higher risks for the health of peri-implant tissues [19‐22]. The cause of this might relate to the fact that the connective tissue barrier surrounding implant-supported prostheses has weak mechanical resistance in comparison to the natural teeth [23‐25]. However, there seems to be a lack of concrete evidence for the exclusion from use of either retention type. Therefore, selection of screw or cement retention is a subject of the patient’s and the dentist’s preference [26, 27]. There are situations that cement retention is a preferable treatment choice for implant-supported crowns, despite its disadvantages. Cement retention might be selected due to esthetics, passivity of fit, occlusal management, unfavorable implant inclination, less cost, and familiarity with the procedures [28‐30]. A plethora of methods have been used for the assessment of the fit of prostheses on abutments [14, 31‐34]. One of the most widely used methods is the replica technique (RT), and according to which, low viscosity light-body silicone material is applied in the crown, which is then seated on the abutment simulating the cementation procedure [14]. Higher-viscosity silicone is then poured in the crown to stabilize the first thin silicone layer. After setting, the silicone replica can be sectioned and measured [33]. This straightforward and inexpensive method has been proven a reliable and accurate alternative to the destructive cross-sectional method [35, 36]. RT has also been modified by making an external impression of the fixed crown on the corresponding abutment and pouring epoxy resin material after setting of the silicone. The marginal fit was then measured on the resin replica [37].

However, the replica technique has limitations [38‐40]. Manual sectioning process of the fragile replica is associated with errors such as inclined cutting, discontinuous cutting, and distortions of the layer. Such errors could give misleading results [2]. Although, cross-sectional images obtained from the slices allow assessment of the prosthesis’ marginal fit, the accuracy of the method depends on the ability of the observer to identify the boundaries between the prosthesis, cement, and abutment as well as to identify the misfit using these boundaries [40]. Furthermore, margins may appear rounded under magnification obstructing an accurate definition of the margin [33]. Also, tearing of the elastomeric film upon removal from the crown has been reported [33, 38]. Measurements of the marginal fit have also been conducted directly on the impression silicone, which functioned as a negative replica (NR) [41, 42]. The NR technique successfully addressed inherent limitations of the RT [38, 39]. Due to study specific equipment, the homogeneity of the replicas was improved, replica slicing procedure was standardized, and sharp digital microphotographs were acquired in a standardized manner. After implementing the digital image analysis sequence (DIAS), a recently developed stepwise procedure, crown margin, abutment finishing line, and the profile of the set cement were recognizable on the two-dimensional image of the replica cross section [41, 42].

The RT has been evaluated in several studies. Despite its limitations, it has been proven for its applicability, repeatability, and validity [2, 43‐46]. However, measurements are often conducted by one examiner [31, 44, 45, 47]. Measurement of agreement is a prerequisite for the acceptance of a new method or process before its application. The agreement between measurements performed by different examiners, reproducibility, is described by an interobserver variation. On the contrary, the closeness between measurements by the same examiner, repeatability, is described by an intraobserver variation [48]. A method with a high degree of reproducibility and repeatability is considered precise and acceptable. A variety of statistical methods has been used to verify agreement, including Pearson’s correlation coefficient, intraclass correlation coefficient (ICC), and non-parametric tests such as Kendall’s tau and Spearman’s rho [48‐53].

Agreement can also be evaluated with the standard error of measurement (SEM), the smallest detectable change at the 95% confidence level (SDC_95%). Furthermore, the Bland and Altman method is a useful tool for the quantification of agreement between examiners by means of graphic representation of the limits of agreement at the 95% confidence level (LoA_95%) [52, 54‐58]. A reasonably large sample size is necessary for statistical tests with an adequate level of power, given the effect size, so that conclusions can be safely drawn when a method is examined [54, 59‐62]. The DIAS is a recently developed and useful method for analyzing MF, while addressing some of the inherent problems encountered with the RT before. Although DIAS is applicable, its validity has not yet been proven. The purpose of this study was to verify the DIAS as an acceptable method for the evaluation of MF by conducting statistical tests of agreement between different examiners using the Bland and Altman method. The hypothesis was that the measurements with DIAS would not be different between and within different examiners.

One hundred thirty digital images were anonymously and randomly analyzed two times by observer 1 and once by each of the three other observers in order to assess the validity of the DIAS. Five groups were derived from this setup with 650 analyzed images in total. The initial MF of the samples in this study ranged between practically ideal fit (4.5 μm) and up to 205 μm. Discrepancies were artificially introduced before cementation with washers of known thickness [6]. The result of Shapiro–Wilk test of normality was P < .05 rejecting the null hypothesis that the data were not different from a normal distribution. The Q-Q plots, histograms, skewness, and kurtosis reflected the same result that the data are not normally distributed and exhibited positive skewness. The descriptive statistics are summarized in Table 1. The median and range are shown first in the table because the median is more representative than the mean when the data are asymmetrically distributed.

Both the intra- and interobserver reliability had excellent ICC scores (ICC ≈ 1, Power (1-β err prob) > 99.99%) (Table 2). The applied alternative methods (non-parametric correlation tests Kendall’s tau (τ) and Spearman’s rho (r_s)) for assessing correlation of measurements show excellent correlation as well (1). Observer 1’s mean measurements of marginal fit implementing DIAS were significantly correlated with observer 2’s measurements (r_s = .998, τ = .977), and observer 3’s measurements (r_s = .998, τ = .971), as well as observer 4’s measurements (r_s = .999, τ = .984) (all Ps < .0001 and Power (1-β err prob) > 99.99%). In Table 3, the calculated SEM and SDC_95% values in pixels are shown.

Tests of normality for the distribution of differences between examiners showed no significant deviations from normal distribution. The results of these tests are summarized in Table 4. Hence, parametric tests were run. The intrarater agreement was determined with the Bland and Altman analysis of bias. The Bland and Altman method included a one-sample t test of differences. The null hypothesis that the set of the measurements of each of the observers 2, 3, and 4 would not be different from the mean measurement of observer 1 was maintained. In the one-sample t test with 129 degrees of freedom, the P > .05 (not significant (ns)) indicated absence of fixed bias. The effect sizes Pearson’s r and Cohen’s d_z were also calculated. On average, the measurements of observer 1 in pixels (M = 117.11, SE =5.61) were not significantly different from those of observer 2 (M = 117.17, SE = 5.58, t = − ,687, P = .494 (ns) > .05, r = .06, d_z = − .06, Power (1-β err prob) = 10.47%), neither from those of observer 3 (M = 117.30, SE = 5.582, t = − 1.648, P = .102 (ns) > .05, r = .114, d_z = − .145, Power (1-β err prob) = 37.29%) nor from those of observer 4 (M = 117.18, SE = 5.61, t = − 1.109, P = .270(ns) > .05, r = .097, d_z = − .097, Power (1-β err prob) = 19.62%).

After calculation of the LoA_95%, Bland–Altman scatter plots were drawn to present graphically the level of agreement between the average measurements of observer 1 and each one of the other observers (2, 3, 4). The Bland–Altman scatter plots of the difference between the first observer’s average measurements and the second (Fig. 4), third (Fig. 5), and fourth (Fig. 6) observer’s measurements were drawn with the aid of SPSS software. The Y-axis illustrates the difference between the first observer’s mean and each other observer respectively (Figs. 4, 5, and 6). The X-axis presents the average of their measures. Difference mean is the estimated bias between the first observer’s and second (Fig. 4), third (Fig. 5), and fourth (Fig. 6) observer’s measurements, respectively. LoA_95% is the limits of agreement at the 95% confidence interval.

This study showed that the DIAS is a reliable and reproducible instrument for the quantitative evaluation of MF on cemented implant-supported crowns. In this study, the followed NR technique and the image acquisition procedure provided a high level of standardization among acquired images. Random subject variation and bias were tackled by blinding the observers with randomly assigning the images for analysis [48]. It was not possible to draw conclusions on the effects of cementation on the MF of the samples with the setup of this study.

After implementing the digital image analysis sequence (DIAS), a recently developed stepwise procedure, crown margin, and abutment finishing line as well as the profile of the set cement were recognizable on the two-dimensional image of the replica cross section [41, 42]. The marginal fit could be examined directly on cemented crowns in situ [8, 9, 34]. However, direct observation of the margin would facilitate only assessment of the vertical discrepancies and only qualitative assessment of the exposed cement surface. Examination of the margin of a cemented crown from the side is a subject of microscope’s depth of field-delicate calibration and orientation of the sample, providing images that seem to be blurry, obscuring accurate measurements. An alternative could be a destructive method with embedding the samples, cross sectioning them and observing the actual sections under the microscope [35]. However, this technique also includes many steps, is time consuming, and might also introduce artifacts at the fragile cement profile. The utilization of a scanning electron microscope would increase costs and would also be time consuming [7]. Utilization of an optical microscope and implementation of sample recycling procedures, which facilitated the reuse of specimens, in the described NR technique minimized costs. Observer subjectivity was minimized with the implementation of strict and clear criteria to identify the cement profile and the MF. Therefore, the within and between observer variation were drastically reduced.

The axial symmetry of the selected two-component samples seems to be a viable model to investigate the abutment-prosthesis system in similar studies. Additionally, the setup of this experiment facilitated standardized procedures. Furthermore, images were checked for artifacts upon acquisition, which minimized observation errors and reduced variation. Moreover, the use of a program to automatically record and organize the data eliminated the human factor from these procedures.

The authors provide a detailed description of the implemented DIAS. MF was defined as the line segment between two easily recognizable and preselected points, located at the marginal edge of each component. Furthermore, each component was accurately identified by manually placing the corresponding outline on the profile of the sectioned replica of the crown-abutment assembly. Manual image analysis procedures might be considered alternative to automated procedures. However, this step was the only source of variability, since the final step was the execution of the measurement in pixels of a well-defined line segment. Although in this study measurements were taken in pixels, ImageJ could be easily calibrated to summarize meaningful units.

In order to ensure reproducibility of results, reliability and agreement were tested with several different methods [52‐54]. The Bland and Altman method facilitated the determination of the systematic error between observer 1 and each of the next three observers in this study, the investigation of the existence of any systematic difference between the measurements, and the calculation of the LoA_95%. The estimated bias was the mean value of the examined difference. The standard deviation of the differences represented the random fluctuations around the estimated bias.

There have been studies focusing on the reliability of RT, comparing it with different methods. However, RT is different from the NR that has been used in this study, hindering comparisons. Additionally, it was not within the scope of this study to compare results deriving from different methods. Since the RT and NR methods share certain similarities such as the idea of a medium that represents the hollow space between fitted components, it is worth mentioning their findings, which are summarized in Table 5.

Furthermore, Mai et al. 2017 studied the agreement between the RT and a computer-aided replica technique utilizing an optical scanner [40]. The agreement estimate for the marginal fit between the two methods was 0.9999, and the precision estimate was 0.9838. Similarly, high scores of correlation statistics were found in this study; however, the context of the correlations was different, prohibiting comparisons with the aforementioned studies. In this study, differences between repeated measurements were not significant, indicating test-retest reliability and low variance of recorded scores. Also, the correlation coefficients scores, for the repeated measurements, revealed that scores between repeated measurements were strongly correlated. Nevertheless, correlation tests are not appropriate to examine agreement between observers, because they cannot reveal possible fixed bias [52].

The experimental error was exceptionally small, increasing confidence in measurements. DIAS demonstrated excellent intra- and interobserver reliability. The ICC values were ≥ 0.99, suggesting excellent reliability [46]. Similarly, high reliability scores have been observed by others that implemented image analysis in their study [44]. Some researchers argue that the SEM is a preferable measure of the quality of an assessment than is reliability, especially in studies with small sample sizes [54]. The SEM depicts the quality of an evaluation free of limitations such as sample size. The SEM both intra- and interobserver in this study was exceptionally small < 1%, which means that the presented methodology has a very low level of measurement error, increasing therefore the confidence in measurements. Otherwise stated, there is more than 95% certainty that the true value is within the ± 2% of the observed value.

In this study, the intraobserver SDC_95% was 1.67% and interobserver SDC_95% value was 2.07%. The meaning of these values is that in 95% of cases the image analysis method could detect a difference in MF if it was larger than about 2% of the mean MF of the total sample. The minor differences between observer 1 and the other three observers that took part in this study might be attributed to the familiarization with the method. Observer 1 had greater experience with the method in comparison with the other three observers despite the fact that they had received adequate training and conducted several trials before they run the analysis. An additional reason might be the motivation to take accurate measurements, because the suggested method is time consuming, requires commitment and precision.

Due to the very low probability that different observers would perfectly agree and proceed with identical ratings for a certain set of measurements, it was examined if different observers would give significantly different ratings with the DIAS. The Bland and Altman method was applied to estimate the likelihood of each observer to give a different measurement from those of observer 1. This approach is often used to examine the limits of agreement between two different methods or different examiners [52, 53]. In order to calculate with a certain level of precision the limits of agreement between two examiners, a relatively large number of observations is required. The sample size in this study was N = 130, which was adequate and reasonably large according to Bland’s recommendations.

The bias in repeated measurements between examiners did not exceed 0.2% (0.166%), and the paired-samples t test could not reveal any bias (ns) at the studied sample size, although the inability to show that a difference exists at the acquired data is not a proof of absence of such difference. Furthermore, the calculated power of the t tests was small (10.47–37.29%); however, the effect sizes (r and d_z) were also small [61, 62]. A small effect size indicates that if a difference actually exists, it might be revealed in a very large sample size and it would be so small that it would not have any practical meaning eventually. The GPower software was utilized for the post hoc calculation of sample size and power [56, 57]. The calculated sample size for parametric tests with adequate power for such effect sizes was more than 2000 samples.

Observers 2, 3, and 4 had, respectively, one, zero, and four outliers in their measurement differences from observers 1’s averaged measurements. While observers 2 and 4 had outliers in their measurement differences, they also had a better agreement with observers 1’s measurements than what observer 3 did. Additionally, the three Bland and Altman scatter plots show very good agreement with minimal observations laying outside the LoA boundaries. Even differences within the ± 5 pixels range, which is larger than the observed accuracy in this study, would not have any practical significance. When transforming the observed discrepancies in meaningful units (0.65 μm/pixel) at the calibrated images of this study, the ± 5pixels range would represent a maximum discrepancy of about 6 μm, which does not seem to have practical value clinically [11].

This was a laboratory study, conducted in relatively stable environmental conditions, which is not the case when cementation takes place in vivo. The applied DIAS was based on identical and of favorable shape (axial symmetry) implant-components. Therefore, the method might be problematic for fixed prostheses on natural teeth. The analysis was based on two-dimensional observations, posing an additional limitation. The examined DIAS was time consuming and required familiarization of the researcher [37].

Further large-sized clinical trials are needed to verify the results and to evaluate feasibility of application in clinical conditions. Combination of the method with pattern recognition software could reduce the required time for an accurate placement of the outlines on the depicted cemented sample profile.

Since the NR technique is non-destructive, it has the potential to be implemented in both ex vivo and in vivo studies. The NR technique in combination with DIAS has the advantage to provide a precision representation of the prosthesis-abutment assembly, as well as their exposed interface, contributing to the quality control of various steps of prosthesis manufacturing. The achieved marginal fit is one magnitude that can be measured with the combination of these reliable techniques. However, accessing the margin in a manner to partially include the surface of the abutment without traumatizing the delicate peri-implant tissues requires a reproducible method that is yet to be verified clinically.

In clinical situation, a NR might be an impression extending to the abutment surface. Due to the dimension stability of modern impression materials, the acquired replica could be sectioned and studied post hoc without burdening the workflow of a busy practice. Furthermore, image acquisition is a technical procedure which does not require medical personnel. Additionally, the DIAS is also possible to be carried out by a non-medical personnel as did in this study. The combination of NR and DIAS can be a valuable tool to evaluate the achieved marginal fit of cemented crowns and assess the quality of cementation, contributing to the quality improvement of delivered prostheses.

Within the limitations of this study, the presented DIAS has excellent intra- and interobserver reliability, small measurement error and sensitivity level that covers the relevant research needs. Additionally, with the prerequisite of axial symmetry of analyzed components, the method could be applied to at least two designs. Furthermore, the negative replica technique is a highly standardized and cost-effective approach for the evaluation of marginal fit. The DIAS in combination with the NR technique is an objective and reliable method with high level of standardization, able to detect and quantify MF observed in clinical practice.