Accurate evaluation of gingival recession dimensions is an essential part of the diagnosis, treatment planning, and outcome evaluation. In the present study, curvature analysis was used to accurately automate the measurement point selection required for recession depth measurement and analyse the examiners' measurement error. The main source of variability for examiners was CEJ measurement point selection. The automated measurements of recession depth using curvature analysis reduce human variability and increase the precision of measurements.
In the present study, a sample of predominately shallow gingival defects was used to compare approaches due to the inclusion of teeth with identifiable natural CEJ only. Shallow gingival defects presenting with small recession depths are hard to evaluate with a periodontal probe due to mentioned limitations of measurement accuracy [5–7]. However, as already outlined by [10], increased resolution of digital measurements to the nearest 0.01 mm enables evaluation of shallow gingival defects as well. Additionally, the importance of measurement accuracy was also emphasised with a superb illustration of the effect of rounding digital measurements on two important parameters for evaluating the success of different treatment techniques, i.e., percentage of root coverage [34] and percentage of defects with complete root coverage [10, 34].
Manual digital measurements of gingival recession exhibit high variability depicted by Bland-Altman plots. Errors are an inherent part of manual measurements and are unavoidable with human involvement [35]. Therefore, for objective comparison of variability between examiners and the unknown true value of recession depth, an automated curvature-based approach was used as a reference method due to automaticity enabling perfect reproducibility of repeated measurements. Despite excellent ICC values and non-significant mean differences between the approaches, Bland-Altman plots revealed a relatively high variability of recession depth measurements with a 95% limit of agreements range of approximately 1 mm. Despite inexperience in digital analysis, Examiner No. 3, being experienced in clinical diagnostics, exhibited the smallest range of 95% limits of agreement, i.e., 0.57 mm. Obtained variability range is much smaller than evaluation with a periodontal probe, i.e., around 2.5 mm, and similar to evaluation with a digital manual approach, i.e., around 1 mm, supported by our results (Fig. 7) and also found in other studies [3, 4]. In contrast to previous digital studies, the only variable part in the present study was measurement point selection; thus, obtained variability can be attributed solely to the measurement points selection. The digital manual approach utilised in the previous studies [3, 4] measured gingival recessions on the digital models and not on cross-sections, including measuring direction and angle variability, into the comparison. The implementation of the proposed approach is straightforward. It requires only a single 3D data measurement analysis software that is free for non-commercial use.
Importantly, the main source of variability is the CEJ measurement point. Both the manual and the automated approach used the shape of the cross-section for the measurement point selection. The main difference between the approaches was the examiners' subjective bias. Our results revealed that the main variability could be attributed to the position of the CEJ measurement point (Fig. 3a), despite inclusion criteria with visible and identifiable CEJ. Curvature analysis revealed a straighter CEJ profile compared to the GM profile, outlined with smaller absolute curvature values (Fig. 4), resulting in a less distinct shape feature than the gingival margin (Fig. 5). Our findings were supported by scatter plots revealing that absolute curvature values of approximately 1 mm− 1 represent an arbitrary threshold where higher deviations in measurement point position can be observed (Fig. 6). To outline the magnitude of the problem, half of the cases exhibit CEJ with curvature less than 1 mm− 1. Furthermore, a great example is comparing the first and second round of measurements for Examiner No. 1 with larger deviations present at higher absolute curvature values as well in the first round. In contrast, in the second round, a "learning effect" was observed with improving precision over the higher range, but importantly, not below the mentioned threshold.
This study was subject to some limitations. First, either measurement point's true position is impossible to determine; therefore, the trueness of measurements could not be evaluated. Despite the perfect reproducibility of point selection presented automated approach requires initial human input to determine the points of interest. In the present study, the point definition was based on definitions from existing literature. However, with time, both the cement and enamel wears away due to the exposure of tooth's root surface to the oral environment [17], rendering CEJ a questionable landmark. The automated curvature-based approach is applicable beyond the limitations of identifiable CEJ, meaning that when defects or restorations are present, the edges of the defect or restorations can be objectively defined by curvature analysis as well. Thus, in digital analysis, the term and landmark of CEJ might have been redefined to "coronal reference point" in the scope of the periodontal measurements. Additionally, curvature analysis can also aid in describing root surface defects' morphology, opening novel insights into the reconstruction of anatomical CEJ before root coverage [36, 37]. However, further research is required for validation in cases without identifiable natural CEJ.
Second, compared to a whole 3D model, a cross-section represents only one out of many available measuring sites. Further research is required to explore novel possibilities analysing whole 3D models. Nonetheless, cross-sections are widely used to evaluate tissue dynamics in periodontal plastic surgery [34, 38–42] and implantology [43–45], enabling great standardisation regarding the selection of measuring site and direction of measurement. Measurement site selection, e.g., selecting a representative cross-section, is also one of the possible variabilities in the measurement of recession depth. For follow-up measurements measuring site variability was eliminated with superimposition of digital models [11, 39], also stating the importance of superimposition accuracy [46].
Third, while the clinical approach utilizes also the colour properties as the additional visual reference, only shape properties were used in the present study. Despite color acquisition of digital dental models in the study, export of color models for further analysis was unavailable due to software limitations. Exporting of color models has become available only recently for some systems and color models proved to be useful in digital measurements of keratinized tissue width with main emphasis on color difference [47]. Further research could be aimed to test the suitability of colour properties on automated CEJ detection.