National and Kapodistrian University of Athens
Αυτοματοποιημένη μέθοδος εντοπισμού και διάγνωσης μασητικών τερηδονικών βλαβών κατά ICDAS II με την ανάπτυξη ενός συστήματος ανάλυσης ψηφιακών φωτογραφιών των δοντιών
Abstract
dc:descriptionIntroductionLiterature in the last decades suggests that there is a substantial decrease in dental caries world wide. This decrease in caries prevalence is attributed to the wide use of fluoride. This is the main reason that this decrease is associated primarily with the smooth surfaces of the tooth and far less with the occlusal ones. Thus the relative percentage of occlusal caries to smooth surface caries has increased over the years. Occlusal surfaces are more vulnerable because fluoride on one hand has a very limited protective effect on them and on the other the complicated 3D morphology of the pits and fissures and the difficulty to remove dental plaque from them makes the problem even bigger. Plaque accumulation and stains make the diagnosis of early carious lesions even more difficult. This is the reason explaining the diversity in caries diagnosis of occlusal caries among dentists.It is also known that early and accurate diagnosis of carious lesions are very important requirements for a correct decision whether the lesion will be treated preventively or with minimal invasive techniques. This need led research to develop classification systems that will assign the lesions based on optical criteria. Thus, the clinician will have a tool to classify lesions more reliably so he will reduce wrong decisions. The most widely accepted classification system today is the ICDAS IIAt the same time research develops different devices for caries diagnosis that can evaluate differences in optical or physical properties of the affected enamel. Just considering the variety of the proposed systems suggests that up until now there is no device that can reliably identify and classify carious lesions based on ICDAS II system. AimThe aim of this study is to develop an automated tool capable of diagnosing and classifying carious lesions of occlusal surfaces of posterior teeth based on the ICDAS II system without the intervention of the dentist and furthermore, to compare the accuracy of the tool against the decision of two experts and against the histological picture of selected areas. The Automated Caries Diagnosis System (ACDS) is the tool developed for this purpose.Materials and methodsThe algorithm developed for this study has two steps. In the first step the ACDS evaluates the occlusal surfaces using segmentation techniques and identifies areas of interest. The identification of the areas of interest is completed without intervention from the user. These areas are primarily pits and fissures and/or decalcified or stained areas of the tooth. In the second step the system evaluates these areas of interest and using random forest algorithm and classifies them in the 7 ICDAS II categories. Then the decision of the ACDS was compared with the classification of two experts. Finally there were 53 areas that were selected and the histological sections of them were compared with the decision of the ACDS.There were 103 extracted permanent teeth used for the study that were photographed with a digital camera under standardized conditions. These pictures were evaluated by 2 experts based in ICDAS II system. Every area of the occlusal surface was identified separately and received different ICDAS II code if necessary. Then the pictures were evaluated by ACDS system and the results were compared. Subsequently, 18 teeth were selected and were used to evaluate the agreement of the ACDS with the histological picture of the lesions. There 40 sections prepared that included 53 areas of interest that were classified according to the ERK system. ResultsThe first question evaluated by the study was if the ACDS can identify correctly areas on the occlusal surface that need to be evaluated for carious lesions (areas of interest). In the study there were 103 teeth that were photographed by a digital camera. When these pictures were evaluated by the experts they identified 425 areas of interest. ACDS identified 340 περιοχές (true positive). It also identified 85 false negative surfaces and 118 false positive. Out of these 118, 42 were classified by the experts as areas with cavitation (ICDAS ≥3) and 76 as initial lesions (ICDAS 1 or 2). Sensitivity is 0,80 accuracy is 0,80 and precision 0,74. The following step after the identification of the areas of interest was the classification. The ACDS was evaluated in two cases. In the first case the surfaces that were identified as healthy, ICDAS=0, by the experts and the system did not identified them at all were considered correct. In this case the mean value (standard deviation) of sensitivity and specificity for all ICDAS groups was 0,89 (0,12) and 0,97 (0,05) respectively, while accuracy was 0,82 (0,19). Precision was 0,83 and coefficient kappa 0,78. The ACDS outcome was also evaluated with the consideration that the healthy areas were missed by the ACDS are wrong. In this case the mean value (standard deviation) of sensitivity and specificity for all ICDAS groups was 0,80 (0,18) and 0,96 (0,04) respectively while accuracy was 0,79 (0,17). Precision was 0,79 and coefficient kappa 0,74. ICDAS II system was designed to classify areas based on the stage of progression of the lesion based on optical criteria but also based on this classification to assist the decision of the management of the lesion. From a clinical point of view it would be important to classify the lesions in 3 categories based on the treatment of the lesion. These 3 categories were healthy surfaces (A) with ICDAS II=0, initial lesions that need preventive treatment (B) with ICDAS 1 or 2 and small of extensive cavitated lesions that need restoration (C) with ICDAS ≥3.Following the same thinking as in the previous section the results were evaluated considering areas identified as health by the experts that they were not identified by the ACDS as correct. The mean value (standard deviation) of sensitivity and specificity for all ICDAS groups was 0,89 (0,11) and 0,93 (0,08) while accuracy and precision and coefficient kappa were 0,88 (0,16), 0,86 and 0,78 respectively.When the results were evaluated with the consideration that healthy areas that were not identified by the ACDS were accounted as wrong the mean value (standard deviation) of sensitivity and specificity for all ICDAS groups was 0,86 (0,12) and 0,92 (0,06) while accuracy and precision and coefficient kappa were 0,82 (0,12), 0,83 και 0,78 respectively. In order to evaluate the histological sections of the teeth the diagnostic cut offs ERK and ERK3 were used. The ERK 2 cut off includes areas with ERK≥2 and ERK 3 includes areas with ERK ≥3. These cut off points were evaluated as the front of the caries in both categories are deep beyond the enamel-dentine interface between out and middle third of the dentine. In the same manner, ICDAS2 and ICDAS3 cut offs were considered areas with ICDAS≥2 and ≥3 respectively. Therefore there were 4 comparisons made. The most balanced results were associated with ERK3 and ICDAS 3 cut offs. In this comparison agreement was 92,5%, sensitivity and specificity were 100% και 91,1% respectively while coefficient kappa was 0,76. Similar agreement was found when histological segments were compared with the optical examination of the experts. The combination ERK3 with ICDAS3 showed sensitivity and specificity 100% and 95,6% respectively, while agreement was 96,2% coefficient kappa was 0,87.Conclusions1.ACDS is the first automated system that evaluates digital pictures of occlusal surfaces and not only identifies areas with possible carious lesions but also classifies them based on the ICDAS II system. 2.The system does not need the participation of the user in any part of the process providing an objective classification of the lesions eliminating the subjectivity of the human factor. 3.When ACDS classifies lesions in the 7 ICDAS II sensitivity is 0,80, specificity 0,96, accuracy 0,80 and precision 0,80 and coefficient k=0,74.4.When ACDS classifies lesions in 3 categories (ICDAS=0, ICDAS=1 or 2 and ICDAS≥3) sensitivity is 0,89, specificity 0,93, accuracy 0,88 and precision 0,86 and coefficient k=0,78.5.ASDC classifies different areas of the occlusal surface differently allowing different therapeutic approach based on the condition of the surface.6.ACDS can incorporate easily and efficiently new knowledge regarding the detection and classification of carious lesions.7.ACDS can be trained so if we create a database with pictures of occlusal surfaces with known ICDAS II classification the performance of the system will improve.Topics for future studies1.Compose a database with pictures of occlusal surfaces that will have a known ICDAS II classification and they will be used to train the ACDS and evaluate the effect of this training in the decision outcome of the system.2.Study the effect on the outcome of the ACDS of the combined evaluation of pictures of occlusal surfaces in visible light and in the infrared spectrum.3.Study of the outcome of the ACDS when the system evaluates pictures taken in vivo from an intraoral camera.
Degree
thesis:*- Grantor dc:publisher
- National and Kapodistrian University of Athens
- Year dc:date
- 2016
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Berdouses, Elias
- Μπερδούσης, Ηλίας
Subjects
dc:subject × 9Rights
- Language dc:language
- gre
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/37318