Abstract
Objective
The main objective of this study was to evaluate the accuracy of optical coherence tomography (OCT) in detecting naturally occurring non-cavitated fissure caries (NCFC) in totality and at different loci by visually assessing cross-sectional OCT scans (B-scan) with an interpretation criterion. The secondary objective was to evaluate the agreement between dimensions of NCFC measured with OCT and polarized light microscopy (PLM).
Methods
71 investigation sites of sound fissure and naturally occurring NCFC on human extracted premolars were identified and scanned with a swept-source OCT. The teeth were then sectioned bucco-lingually at the investigation sites and imaged using PLM. Two calibrated examiners trained on the B-scan NCFC visual interpretation criteria established for this study, assessed the investigation sites and results were validated against PLM.
Results
Detection sensitivity of B-scan for NCFC when fissures were assessed in totality, or on the slopes or walls separately are 0.98, 0.95, 0.94 and specificity are 0.95, 0.90, and 0.95. One-way ANOVA showed that width measurements of wall loci done with OCT and PLM were not statistically different. However, OCT height measurements of slope loci were statistically bigger with a constant bias of 0.08 mm (of which is not clinically significant) and OCT height measurements of wall loci were statistically smaller (0.57 mm) and Bland-Altman plots indicated presence of proportionate bias.
Conclusion
Visual assessment of B-scans with the interpretation criteria resulted in both high specificity and sensitivity and were not affected by loci location. OCT width measurement of wall loci is in agreement with PLM.
Clinical significance
Unanimous high sensitivity in this and previous studies indicate that visual assessment of B-scans reliably rule out NCFC. Detection accuracy was not affected by loci location. Width of wall loci and/or height of slope loci in OCT B-scan are to be used for monitoring NCFC but not height of wall loci.
1
Introduction
3.9 billion people are affected with oral conditions globally amongst which, untreated caries in permanent teeth is of the highest prevalence [ ], with approximately 35.3 percent of adults suffering from this condition. Although occlusal surfaces represent only 13 percent of total surfaces of the permanent dentition [ ], 88% of caries experienced by schoolchildren in the United States has been found to be pit and fissure lesions [ ] and pit and fissures lesions had also been found to comprise the majority of the caries increment observed in both fluoridated and non-fluoridated communities [ ]. Hence its prevention via fissure sealant in large scale public health effort had been advocated and carried out. Recent studies however had showed that unjudicial application of fissure sealant, especially in population with low caries risk does not increase the prevention potential of fissure sealant. Risk-adjusted remineralisation therapy for incipient fissure caries with fluoride, calcium-phosphate products and bioglass had been proposed and are being explored. Hence the detection and monitoring of progression of incipient lesions has become an integral part of modern caries management [ ]. Incipient lesions had been defined as lesions with surface that appear macroscopically intact with no visual evidence of cavitation [ ]. Such incipient lesions has also been referred to as non-cavitated caries lesions [ ] and those along fissures can be termed as non-cavitated fissure caries (NCFC).
Visual inspection and radiography are the most ubiquitous methods used in the detection of fissure caries lesions clinically. Visual detection system such as the International Caries Detection and Assessment System (ICDAS) has been proven to have good accuracy and reproducibility [ ]. However, such systems require training and calibration with a reference examiner [ ] and can be time consuming [ ]. It also falls short of true quantification as it uses qualitative assessment of features such as colour, texture and roughness. Radiograph has been proven to be useful in detecting occlusal fissure lesions that had extended into dentine. However, the sensitivity of detecting occlusal fissure lesions which is limited in enamel is low (0.14–0.38) [ ]. Therefore, supplementary detection tools such as those based on laser fluorescence and electrical conductance had been explored and used for the detection of NCFC. However, they still lack either sensitivity or specificity [ ].
Optical coherence tomography (OCT) is a method that can obtain cross-section images non-invasively, by measuring the backscattered intensity in depth of a specimen. Changes in the optical scattering properties of the enamel and dentine surfaces have been shown to occur when they are demineralised and a strong correlation between mineral loss and increased scattering coefficient have been observed [ ]. This correlation of reflectivity and demineralisation has been used both qualitatively [ ] and quantitatively [ ] to either detect or quantify the severity and depth of a caries lesion. Quantitative methods attempt to use mathematical algorithm to automate analysis of the depth-profile of reflectivity (A-scans) for severity, depth and structural features of caries lesion. However, there are still challenges faced by these automated approaches such as the selection of cut-off points [ ] or validation of edge detection algorithms and distinguishing artefacts from lesions [ ] and reported opposing interpretations of A-scan profiles may suggest that it could be an unreliable indicator of lesion severity [ ].
Moreover, fissure caries are often multi-loci. It is most frequently localized in the lower part of the walls of the fissures (61%) whilst 36% had multiple foci at the upper and lower part of the fissure and 13% were positioned above the entrance of the fissure [ ]. The angulation and the distance of the area above the entrance of the fissure (slope) and the wall of the fissure, relative to source of the OCT light is substantially different. In view of that, the nature of light attenuation and therefore the detection performance could be affected by whether the foci of the lesion were on the slope of a fissure or at the bottom of wall of a fissure. A recent study concluded that different incidence angle of up to 40° affects significantly the detection of fissure caries [ ] with angulation other than 0° showing better detection performance. The development of automated algorithm has so far been developed and investigated on solitary lesions [ ] and with an incidence angle being 0°. The applicability of these algorithm on fissure caries lesion with multiple loci on the slopes and walls of the fissure is hence unknown. It would also involve more volume of data and potentially require higher computational power and time to analyse the multiple loci.
For a detection method to be successfully translated into the clinical setting, it needs to be relatively economical, does not need to be scanned at multiple axis so as not to be confounded by patient movements, and quick and accurate interpretation of the presence or absence of a non-cavitated lesion. Qualitative assessments are carried out by visually assessing an OCT B-scan to determine the presence of areas of increased backscattered intensity, which is consistent to a caries lesion. This is analogous to the determination of presence of radiolucency of dentine caries lesions in bitewing radiographs. Qualitative assessment of B-scans, although only provides a dichotomous result of presence or absence of caries, it is nevertheless quick and of good practicality. This is especially true in the simultaneous appreciation of the distribution, pattern and dimension of the multiple loci of fissure caries lesions.
However, strong surface specular reflectivity of enamel confounds the reflectivity increase caused by demineralisation and hence produces false positive results. Polarisation-sensitive (PS-)OCT has been shown to be able to eliminate this confounding effect as the perpendicular axis PS-OCT B-scans were found to provide better resolution of the lesion structure including the respective surface zones [ ]. Cross-polarisation (CP-) OCT on the other hand has been shown to reduce specular reflectance by 20–30 dB and the reflectivity from sound enamel is also suppressed near the tooth surface resulting in higher contrast between sound and demineralised enamel [ ].With Swept Source (SS-) and Spectral Domain (SD-) OCT (hereafter to be referred collectively as OCT), techniques such as digital image processing [ ], creating an angle up to 45° to the incidence of light [ ], use of index matching media [ ], systematically excluding combination of depths [ ], applying threshold level [ ] and a statistical cut-off depth to data [ ] are a few examples used to reduce the specular reflection in OCT. Such methods are useful on smooth surface caries lesions with relatively homogeneous demineralisation but are not feasible on occlusal surface with convoluted geometry and multi-loci fissure caries. The confounding effect of surface specular reflection is reflected in the high sensitivity but varied specificity reported in previous studies that involved one rater. Gomez et al. [ ] reported high sensitivity (0.98) but poor specificity (0.52) of OCT in detecting fissure caries whilst Shimada et al. [ ] reported similar sensitivity of 0.98 but a higher specificity of 0.75. When more than one rater were involved, huge inter-rater kappa range (0.25–0.75)[ ] was reported. It is not clear what visual interpretation criteria, if any, was used in these studies and more importantly whether there was any systematic attempt to exclude specular reflection from the interpretation. It is also not clear whether only the wall or slope loci or both of NCFC were considered because it is evident from Park et al’s study [ ] that the inclusion or exclusion of the wall or slope lesions would affect its detection performance.
Measurement of lesion dimension could be used to monitor progression or regression of caries lesions over time. Remineralisation of enamel caries lesions has been shown to result not only in the decrease of backscattered intensity but also the decrease of the area with observed increased reflectivity [ ]. Fried et al. conducted a longitudinal study of simulated-caries lesion using pH cycling model over 1 and 14 days and concluded that depth and severity of lesion is linearly correlated with the square root of demineralisation and PS-OCT is valuable to observe the course of lesion advancement over time [ ]. However, the accuracy of the measurement is not known.
Thus, the main objective of our study is to evaluate the detection accuracy of OCT for NCFC when it is assessed in totality and when only the wall or slope loci was assessed using a visual interpretation criteria. The secondary objective was to evaluate the agreement between the measurements of the wall and slope loci of NCFC measured with OCT with that measured with Polarized Light Microscopy (PLM). The null hypothesis is there is no significant difference between PLM and OCT measurements.
2
Materials and methods
2.1
Sample preparation
Fifty-four extracted permanent maxillary and mandibular human premolars were used in this study. Teeth with fluorosis and staining were excluded from the study. The teeth were stored in 1% chloramine solution for 2 weeks to inhibit bacterial growth and in normal saline thereafter. The fissures were cleaned with slurry of pumice with polishing tips with a low speed hand piece and rinsed with water for 10 s to ensure removal of any residuals. The teeth were then sectioned 3 mm below the cemento-enamel junction and mounted in epoxy resin leaving the crowns exposed. In order to ensure equal distribution of NCFC of different stages, ninety preliminary investigation sites with ICDAS code 0, 1, 2 (n = 30 each) were identified using standard ICDAS protocol ( http://www.icdas.org/ ) by two trained and calibrated examiners (EZ and SK).
2.2
OCT imaging of fissure
The premolars were kept moistened, and just before being scanned with OCT, they were air dried for 5 s with a three-in-one syringe. A Swept-source OCT Imaging System (OCS1300SS, Thorlabs Ltd., UK) with emission wavelength centred at 1325 nm and axial and transverse resolution of 9 μm and 11 μm in air, respectively, was used to scan the identified invertigation sites. The Thorlabs OCT capturing software (Swept Source OCT Imaging System Version 2.3.1, Thorlabs) was used to capture the image, configure the OCT settings and guide the light beam. The scanning beam was oriented perpendicular to the fissure and configured to cover an area of 3 mm x 3 mm in the x-y direction and a depth of 3 mm in air, corresponding to 1.85 mm in enamel (refractive index = 1.62). The x-y-z resolution was 1024 × 104 × 512 pixels, respectively. Therefore, a total of 104 OCT cross-sectional scans (B-scans), approximately 30 μm apart, were attained for each investigation site in order to facilitate the identification of a best matching OCT B-scan to the PLM section. The B-scans were captured in logarithmic scale and saved in the Large DR3 colour map of the image capturing software. The brightness and contrast were configured to cover an intensity range between −36 dB and −10 dB.
2.3
Polarized light microscopy (PLM)
Two 180 ± 10 μm thick-serial sections, 200 ± 10 μm apart, were obtained from each of the identified investigation site using the Exakt 300CP Band system (Exakt, Germany). The sections were made in the bucco-lingual direction perpendicular to the fissure using a cutting band of 200 μm thickness with grade D diamonds. No further polishing and finishing was performed after sectioning to avoid affecting the lesions. The section with the most severe lesions of the two was used. The selected sections were then imbibed in water and examined under 4×-magnification. The microscope (Nikon Eclipse 55i, Nikon, Japan) was equipped with an analyzer slide (Nikon C-AS Nikon, Japan), a polariser (Nikon C-TP Nikon, Japan) for first order red compensation, and an integrated digital camera (Nikon DSFI2, Nikon, Japan).
2.4
Final sample selection
From the PLM sections, a final cohort of seventy one investigation sites were selected based on the Ekstrand histology criteria [ ] ( Table 1 ), with a distribution of Ekstrand score 0 (n = 21), 1 (n = 25) and 2 (n = 25). The Ekstrand score 1 and 2 were grouped collectively as NCFC for the subsequent analysis. The OCT B-scans that correponded to the selected PLM sections were then identified from the available 104 B-scans, based upon fissure anatomy and cuspal slopes by two examiners (EZ, HPC).
| Score 0 No enamel demineralisation or a narrow surface zone of opacity (edge phenomenon). |
| Score 1 Enamel demineralisation limited to the outer 50% of the enamel layer. |
| Score 2 Demineralisation involving between 50% of the enamel and outer third of the dentine. |
2.5
Interpretation of OCT B-scans and PLM images
2.5.1
Division of fissure
For the interpretation of the OCT B-Scans and PLM images, the fissures were divided into two anatomical sites at the entrance of the fissure, as described first by Fejerskov et al. [ ] and later modified by Juhl [ ]. The entrance of the fissure ( Fig. 1 ) is defined as the point where its width was 200 μm [ ]. In order to facilitate description, the area above the entrance hereafter was termed as slopes ( S ). The limits of the slope was within 500 μm from the entrance of the fissure ( Fig. 1 ) and were labelled on each of the OCT B-scans and PLM images. The area below the entrance of the fissure was termed as walls ( W ) ( Fig. 1 ). Consequently, the lesion foci observed at the slopes were referred as S L and those observed at the walls referred as W L ( Fig. 1 ).
The OCT B-scans and PLM images were displayed on a computer screen and standardised in size, aspect ratio and brightness for evaluation by two trained and calibrated examiners (EZ and MH). Assessment of the fissure for NCFC was performed in separate occasions, for the following sites:
- i)
Fissure in totality ( T ) (including the slopes and walls).
- ii)
Slopes ( S ) only.
- iii)
Walls ( W ) only.
2.5.2
Interpretation criteria for absence and presence of caries
On the PLM images, absence of dark brown and black areas at the investigation sites were considered as sound ( Fig. 2 a), whereas, the presence of these dark brown and black areas were regarded as carious ( Fig. 2 b).
On the OCT B-scans, a NCFC is considered to be present when bands of elevated backscattered intensity, in the range of −15 dB to −5 dB, which are sub-surface and diffused in nature ( Fig. 2 d) were observed. Intensity in the same range which is limited to one to two pixels thick on the surface was NOT considered as carious ( Fig. 2 c). This intensity range is presented as yellow and red by the Large DR3 colour map of the Thorlabs OCT capturing software.
2.6
Measurements of lesion dimension
In addition to the above, measurement of the dimension of the NCFC was carried out with PLM and OCT. The measurements were done on the OCT B-scans and PLM images using the Thorlabs OCT software (Swept Source OCT Imaging System Version 2.3.1, Thorlabs, USA) and the Nikon software (NIS- Elements AR 3.2 Nikon Japan), respectively.
The following dimensions were obtained ( Fig. 1 ) :
- i)
Height of slope loci ( hS L ): The vertical extent of S L , within 500 μm from the entrance of the fissure.
- ii)
Width of wall loci ( wW L ): The horizontal extent W L
- iii)
Height of wall loci ( hW L ): The vertical extent W L
Two trained and calibrated examiners (EZ and MH) measured the NCFCs on OCT B-scans and PLM images separately and on two different occasions. For the assessment of level of agreement between the measurements done with OCT and PLM, the measurements of the two examiners were averaged. A consensual measurement was obtained for inter-examiner measurements that differ greater than 50%. The difference of measurements between these two techniques, Δm (PLM measurement − OCT measurement), was calculated for each sample.
2.7
Statistical analysis
Statistical analysis was performed using SPSS (SPSS version 12, IBM, USA). Inter-examiner agreement for the detection of lesions was calculated using Cohen’s un-weighted kappa. The detection sensitivity and specificity of OCT for NCFC, as a whole or separately on the slope or wall of the fissure were compared against PLM. For the analysis of sensitivity and specificity, a consensus score was sought in the event of a discrepancy between examiners.
Inter-examiner agreement for the measurement of lesion dimension was calculated using single measure intra-class coefficient (ICC), with a 2-way mixed effect model and consistency type. One-Sample T test was performed to evaluate whether Δm was significantly different from zero. The measurement agreement between OCT and PLM was also evaluated using the Bland-Altman plot. In cases where Δm was significantly different from zero, linear regression was carried out on the Bland-Altman plot to evaluate the existence of bias.
2
Materials and methods
2.1
Sample preparation
Fifty-four extracted permanent maxillary and mandibular human premolars were used in this study. Teeth with fluorosis and staining were excluded from the study. The teeth were stored in 1% chloramine solution for 2 weeks to inhibit bacterial growth and in normal saline thereafter. The fissures were cleaned with slurry of pumice with polishing tips with a low speed hand piece and rinsed with water for 10 s to ensure removal of any residuals. The teeth were then sectioned 3 mm below the cemento-enamel junction and mounted in epoxy resin leaving the crowns exposed. In order to ensure equal distribution of NCFC of different stages, ninety preliminary investigation sites with ICDAS code 0, 1, 2 (n = 30 each) were identified using standard ICDAS protocol ( http://www.icdas.org/ ) by two trained and calibrated examiners (EZ and SK).
2.2
OCT imaging of fissure
The premolars were kept moistened, and just before being scanned with OCT, they were air dried for 5 s with a three-in-one syringe. A Swept-source OCT Imaging System (OCS1300SS, Thorlabs Ltd., UK) with emission wavelength centred at 1325 nm and axial and transverse resolution of 9 μm and 11 μm in air, respectively, was used to scan the identified invertigation sites. The Thorlabs OCT capturing software (Swept Source OCT Imaging System Version 2.3.1, Thorlabs) was used to capture the image, configure the OCT settings and guide the light beam. The scanning beam was oriented perpendicular to the fissure and configured to cover an area of 3 mm x 3 mm in the x-y direction and a depth of 3 mm in air, corresponding to 1.85 mm in enamel (refractive index = 1.62). The x-y-z resolution was 1024 × 104 × 512 pixels, respectively. Therefore, a total of 104 OCT cross-sectional scans (B-scans), approximately 30 μm apart, were attained for each investigation site in order to facilitate the identification of a best matching OCT B-scan to the PLM section. The B-scans were captured in logarithmic scale and saved in the Large DR3 colour map of the image capturing software. The brightness and contrast were configured to cover an intensity range between −36 dB and −10 dB.
2.3
Polarized light microscopy (PLM)
Two 180 ± 10 μm thick-serial sections, 200 ± 10 μm apart, were obtained from each of the identified investigation site using the Exakt 300CP Band system (Exakt, Germany). The sections were made in the bucco-lingual direction perpendicular to the fissure using a cutting band of 200 μm thickness with grade D diamonds. No further polishing and finishing was performed after sectioning to avoid affecting the lesions. The section with the most severe lesions of the two was used. The selected sections were then imbibed in water and examined under 4×-magnification. The microscope (Nikon Eclipse 55i, Nikon, Japan) was equipped with an analyzer slide (Nikon C-AS Nikon, Japan), a polariser (Nikon C-TP Nikon, Japan) for first order red compensation, and an integrated digital camera (Nikon DSFI2, Nikon, Japan).
2.4
Final sample selection
From the PLM sections, a final cohort of seventy one investigation sites were selected based on the Ekstrand histology criteria [ ] ( Table 1 ), with a distribution of Ekstrand score 0 (n = 21), 1 (n = 25) and 2 (n = 25). The Ekstrand score 1 and 2 were grouped collectively as NCFC for the subsequent analysis. The OCT B-scans that correponded to the selected PLM sections were then identified from the available 104 B-scans, based upon fissure anatomy and cuspal slopes by two examiners (EZ, HPC).
