Deep-learning-based automated detection and assistance in the diagnosis of early dental carious lesions

doi:10.21203/rs.3.rs-3125352/v1

Download PDF

Research Article

Deep-learning-based automated detection and assistance in the diagnosis of early dental carious lesions

https://doi.org/10.21203/rs.3.rs-3125352/v1

This work is licensed under a CC BY 4.0 License

Journal Publication

published 31 Oct, 2023

Read the published version in Clinical Oral Investigations →

You are reading this latest preprint version

Objective

To investigate the effectiveness of a convolutional neural network (CNN) in the detection of healthy teeth and early carious lesions on occlusal surfaces, and to assess the applicability of this deep-learning algorithm as an aid in the diagnosis of dental caries.

Materials and Methods

A total of 2,481 posterior teeth (2,459 permanent and 22 deciduous teeth) with varying stages of carious lesions were classified according to the International Caries Detection and Assessment System (ICDAS). After clinical evaluation, ICDAS 0 and 2 occlusal surfaces were photographed with a professional digital camera. VGG-19 was chosen as the CNN and the findings were compared with those of a reference examiner to evaluate its detection efficiency. To verify the effectiveness of the CNN as a diagnostic aid, three examiners (an undergraduate student (US), a newly graduated dental surgeon (ND), and a specialist in pediatric dentistry (SP) assessed the acquired images (Phase I). In Phase II, the examiners reassessed the same images using the CNN-generated algorithms.

Results

The training dataset consisted of 8,749 images, whereas the test dataset included 140 images. VGG-19 achieved an accuracy of 0.879, sensitivity of 0.827, precision of 0.949, and F1-score of 0.887. In Phase I, the accuracy rates for examiners US, ND, and SP were 0.543, 0.771, and 0.807, respectively. In Phase II, the accuracy rates improved to 0.679, 0.886, and 0.857 for the respective examiners. The number of correct answers was significantly higher in Phase II than in Phase I for all examiners (McNemar test;P < 0.05).

Conclusions

VGG-19 demonstrated satisfactory performance in the detection of early carious lesions and as a diagnostic aid.

Clinical relevance:

Automated detection of early carious lesions by deep-learning algorithms is an important aid in the early diagnosis of the disease, as it minimizes subjective assessments by different examiners, enabling quicker and more reliable clinical decision-making.

Deep Learning. Dental Caries. Convolutional Neural Networks. Clinical Decision-making

Dental caries is ranked as the second disease that most widely affects the health of human populations [1], and it has a multifactorial and dynamic etiology [2]. The development of carious lesions results from an imbalance between the demineralization and remineralization of dental hard tissues, influenced by the frequent consumption of high-sugar foods and poor oral hygiene practices [3]. As carious lesions do not regress naturally and are not effectively treated with short-term pharmacological interventions, they can negatively affect individuals’ quality of life [4–10] and become the major cause of dental pain [5, 11, 12].

In the clinical setting, the diagnosis of dental caries can be challenging for clinicians due to the diverse characteristics of lesions at different stages of development; therefore, they require distinct treatment approaches [13]. Early lesions, known as white spots, are reversible and can be managed through noninvasive interventions such as fluoride therapy in association with control over sugar consumption [14]. Cavities, on the other hand, involve loss of the tooth structure and require moderately to highly complex restorative procedures, depending on the depth and severity of the lesion [15].

Recognizing and detecting the various stages of carious lesions, mainly the early ones, is crucial in this context, as it allows for early disease control and the implementation of treatment strategies using minimally invasive procedures [16] in order to preserve healthy tooth structures to the greatest extent possible [3, 17]. Also, detection of early carious lesions enables more conservative and cost-effective treatments, avoiding repetitive and complex treatments that could eventually lead to tooth loss [17].

Visual examination conducted in the dental office under artificial lighting with the aid of a blunt-tipped probe is still the main diagnostic method for the detection of carious lesions [13]. This reliable, noninvasive, cost-effective, and easily performed method allows detecting the initial stages of carious lesions [18]. Moreover, unlike other diagnostic methods, visual examination permits evaluating the activity, depth, and retention potential of carious lesions [15]. Some detection systems have been devised to preserve the sensitivity and high precision of this method, thus reducing the influence of the clinician’s experience during the procedure [19, 20]. The International Caries Detection and Assessment System (ICDAS), proposed by Ismail [21], has been widely used because it allows for the measurement of healthy and carious surfaces, including early lesions, encompassing multiple clinical aspects and providing a comprehensive assessment of tooth condition. Studies have shown that ICDAS has a diagnostic accuracy comparable to that of histological analysis in dental caries detection [22, 23]. Nevertheless, the accuracy of ICDAS relies directly on the clinician’s expertise and previous training [21].

Numerous auxiliary exams, such as radiographs, computed tomographic scans, transillumination, and histological analysis, can help detect carious lesions [22]. These methods, however, have major limitations for their clinical application and they show variable effectiveness in the detection of carious lesions, especially in earlier stages [25]. Hence, in recent years, there has been a growing demand for automated methods, such as convolutional neural networks (CNNs), which leverage deep-learning techniques to assist clinicians in detecting dental caries in the very first clinical signs [26–29].

CNNs are trainable algorithms capable of automating the detection of patterns and classifying changes by extracting data, such as shape, illumination, and color distribution, from images [30]. A large number of dental images can be utilized for CNN training, including radiographs [31, 32] and infrared transillumination images [33, 34], but intraoral photographs, regarded as the best diagnostic method for the detection of early carious lesions, closely resemble visual examination [35–42].

Assessing the effectiveness of CNN training based on dental images for the detection of dental caries is still incipient in dentistry, and to the best of our knowledge, only eight studies have addressed it [35–42] and only two studies have assessed the effectiveness of neural networks in the detection of early lesions [37, 38]. Despite the growing interest of researchers in automated detection of carious lesions, no studies have tested the use of CNN as a diagnostic aid by examiners with different levels of clinical experience. Therefore, the effectiveness of CNNs in dentistry should be further investigated by training a network capable of detecting early carious lesions and assisting clinicians in their diagnosis.

The aim of this study was to assess the application of CNNs in the detection and classification of health teeth and early carious lesions using occlusal images of extracted teeth, and to verify the effectiveness of neural networks as an auxiliary tool for the diagnosis of dental caries.

Ethical aspects

The study protocol was submitted to and approved by the Human Research Ethics Committee of the Division of Health Sciences of the Universidade Federal do Paraná (UFPR) (CAAE 25001219.5.0000.0102). This study follows the Standards for Reporting of Diagnostic Accuracy Studies (STARD) [43] and Checklist for Artificial Intelligence in Medical Imaging (CLAIM) [44].

Sampling

Approximately 6,800 extracted deciduous and permanent posterior teeth were assessed. The teeth were provided by the Human Teeth Bank (HTB) of the Dental School of UFPR and by the Division of Biological Sciences of the Universidade Estadual de Ponta Grossa (UEPG). Non-carious teeth with extrinsic or intrinsic stains, or those which had undergone restorative procedures (e.g., sealants, restorations and denture placement), or endodontic treatments such as crown opening and visible filling material were excluded. Teeth with developmental defects of enamel were included providing there was no loss of the tooth structure that would hinder clinical examination of the carious lesions. In this study, 2,481 teeth (HTB-UFPR: 2.311; HTB-UEPG: 170) were selected, of which 2,459 were permanent and 22 were deciduous, encompassing different stages of dental carious lesions.

Clinical examination

The extracted teeth were examined under artificial lighting with the aid of a LED flashlight strapped to the evaluator’s head and a blunt-tipped probe [13]. All teeth were identified by a number (ID) and assessed for the presence of dental carious lesions, using ICDAS II codes [21]. The occlusal surface of each tooth was given an ICDAS score between 0 and 6, where 0 – Slight or no change in enamel translucency after drying for 5 s; 1 – Hardly visible opacity on the wet surface, but noticeable after drying; 2 – Visible opacity without drying; 3 – Cavitation on the opaque or pigmented enamel, without visible dentin; 4 – Shading of underlying dentin, with or without enamel rupture; 5 – Cavitation on opaque or pigmented enamel with exposure of the underlying dentin; 6 – Cavitation on opaque or pigmented enamel with exposure of the underlying dentin, involving more than 50% of the surface [21]. In this study, scores “1” and “2” were combined because of detection limitations on wet surfaces of the extracted teeth. The clinical examination was performed by three previously calibrated evaluators, who underwent theoretical and practical training and had satisfactory results for all clinical aspects assessed: intrarater reliability (kappa statistic ≥ 0.889 and confidence interval > 95%) and interrater reliability (K ≥ 0.8133 and CI > 95%), which were regarded as optimal agreement [45]. The final classification of the index was made by a fourth evaluator (reference examiner). The data were tabulated using SPSS (version 20.0 IBM, Brazil).

Image databases

After clinical examination, extracted teeth whose occlusal surfaces had a score of “0” (ICDAS-0) and “2” (ICDAS 1 and 2) were selected. The images were photographed by a trained evaluator, using a Canos EOS Rebel T61® camera equipped with a Canon Macro Ring Lite MR-14ex® flash and a macro Ultrasonic® lens. The images were standardized by configuring the camera as follows: shutter speed 1/100, f22 aperture, and ISO 100. Multiple images of the same tooth were captured to simulate variations in clinical visualization, altering the angulation and focal distance. High-resolution photographs were saved as JPG files, with dimensions of 6000x4000 pixels and vertical/horizontal resolution of 72 dpi.

Prior to constructing the image database for CNN learning, the occlusal images of the extracted teeth underwent pre-processing. First, the photographs were labeled by three trained evaluators using the Label Image Regions software from the Scikit-image library, developed in Python [46], and saved in Pascal/VOC format. After labeling, the teeth were cut out of the images and the background was removed from each image, resulting in image blocks that solely contained the occlusal surface of the extracted tooth. Figure 1 provides examples of labeled and cropped photographs of ICDAS-0 and ICDAS-2 teeth.

Deep-learning model

The CNN VGG-19 algorithm, developed by Karen Simonyan and Andrew Zisserman [47], was used in the present study. This model is well suited for tasks such as object detection, classification, and localization within image datasets. The input data of the network are preprocessed and the mean RGB value calculated from the training set is then subtracted from each pixel [47]. The process includes convolutional layers with 3x3 or 1x1 filter sizes and convolutional stride of 1 pixel, covering the entire image. The max-pooling operation is performed in a 2x2 pixel window, with a stride of 2 [47].

VGG-19 training and test

The images were split into training and test datasets, allowing the algorithm to achieve strong generalization capabilities. The utilized VGG19 model processed 256x256 colored images, resulting in a matrix of dimensions (256,256,3). A learning rate of 10^− 6 (0.000001) was employed, and transfer learning was applied using network weights previously trained in the imagenet database [47, 48]. The training consisted of 30 sessions. The classification structure of the network was modified. The three final dense layers, with sizes of 4096, 4096, and 1000, were removed and replaced with six layers of sizes 1000, 800, 400, 200, 100, and 2. The final layer, with size 2, corresponds to ICDAS-0 or ICDAS-2. The data augmentation technique was used during the training phase, and all training images went through some changes (rotation, horizontal and vertical flipping, shearing, and addition of Gaussian noise) for the creation of new synthetic images. Also in this stage, VGG-19 was tested in a validation subset at each iteration. The whole training process lasted on average around 101 min.

After the training, the classifier was tested in the test dataset and the results obtained from the network were compared with those of the reference examiner to obtain the VGG-19 performance metrics: accuracy, sensitivity, precision, and F1-score. The optimal number of epochs was 25. The results described in what follows are those obtained for this epoch.

During the testing phase, the clinical applicability of the CNN as an aid in diagnosing early dental carious lesions was assessed. In Phase I, the images in the test dataset were assessed by three evaluators: an undergraduate student (US), a newly graduated dental surgeon (ND), and a specialist in pediatric dentistry (SP). Each examiner assessed the occlusal images individually and in a blinded mode, giving each assessed tooth an ICDAS score “0” or “2”. The scores were entered into an online table in Google Sheets. After seven days, the same examiners reassessed the images (Phase II), taking into consideration the results obtained from the classifier. They then decided whether to keep or change the scores given in Phase I. Performance metrics were obtained for Phases I and II, and the difference in the number of correct answers between the two phases was statistically analyzed by McNemar test, using STATA version 12.0 (StataCorp LP, College Station, USA). The significance level was set at 5%.

Database characteristics

A total of 1,778 occlusal photographs were taken of the extracted teeth. After application of the data augmentation technique, the training dataset was expanded to include 8,549 images. A total of 200 images were selected for validation purposes. After the training phase, the classifier was tested in the test dataset, which consisted of 140 images of ICDAS-0 (n = 70) and ICDAS-2 (n = 70) teeth. The validation and test datasets were not subjected to data augmentation and were carefully organized to ensure there were no duplicate images in the training dataset. The database used for network learning and testing comprised a total of 8,889 images (Table 1).

Table 1

Distribution of occlusal photographs of extracted teeth in the training/validation and CNN test datasets, according to ICDAS criteria.
ICDAS	Training dataset			Test dataset	Total datasets
ICDAS	Training	Validation	Total	Test dataset	Total datasets
0	4,543	100	4,643	70	4,713
2	4,006	100	4,106	70	4,176
Total	8,549	200	8,749	140	8,889
ICDAS = International Caries Detection Assessment System

VGG-19 performance metrics

In the test dataset, the network demonstrated an accuracy of 0.879, a sensitivity of 0.827, a precision of 0.949, and an F1-score of 0.887 for the detection of oral health patterns and early carious lesions on the 140 images of the extracted teeth (Table 2).

Table 2

VGG-19 metrics for detection of early carious lesions.
Parameter
Accuracy	0.879
Sensitivity	0.827
Precision	0.949
F1-score	0.887

Diagnostic aid

Examiners US, ND, and SP obtained 76, 108, and 113 correct answers, respectively, in Phase I, and 95, 124, and 120 correct answers, respectively, in Phase II. The number of correct answers was significantly higher when comparing Phases I and II for all participants (p < 0.05). The accuracy of the examiners in Phase I was as follows: US = 0.543, ND = 0.771, and SP = 0.807. In Phase II, the examiners achieved the following accuracy: US = 0.679, ND = 0.886, and SP = 0.857. With respect to sensitivity and precision, the examiners obtained US = 0.54 and 0.54; ND = 0.88 and 0.65; SP = 0.96 and 0.66 in Phase I, and US = 0.76 and 0.88; ND = 1.0 and 0.77; SP = 0.97 and 0.74 in Phase II. The F1-score was 0.54, 0.68, and 0.74 in Phase I and 0.68, 0.87, and 0.83 in Phase II for US, ND, and SP, respectively (Table 3).

Table 3

Performance of participants in the detection of early dental carious lesions before and after access to CNN results.
	Undergraduate student		Newly graduate in dentistry		Specialist in pediatric dentistry
Metric	Phase I	Phase II	Phase I	Phase II	Phase I	Phase II
Correct answers	76	95	108	124	113	120
p*	0.001		0.005		0.003
Accuracy	0.543	0.679	0.771	0.886	0.807	0.857
Sensitivity	0.54	0.76	0.88	1.0	0.96	0.97
Precision	0.54	0.60	0.65	0.77	0.66	0.74
F1-score	0.54	0.68	0.74	0.87	0.78	0.83
*McNemar test
Significant results are shown in boldface.

Dental caries is a highly prevalent disease [1] and its early detection can pose some challenges to clinicians [50]. Early diagnosis allows for a more appropriate treatment, preserving the healthy tooth structure as much as possible, thus preventing the further development of lesions and more complex restorative treatments or tooth extraction [17]. Therefore, the development and validation of the applicability of deep-learning as early carious lesions diagnostic aid based on occlusal photographs are crucial to help clinicians make informed decisions and treatment plans. In addition, the CNN can streamline and expedite clinical examination and also minimize subjective judgment during visual examination, which is influenced by the examiner’s expertise and performance. Such advancements can contribute to enhancing patient safety.

VGG-19 was effective in detecting early carious lesions on occlusal images of extracted teeth, exhibiting satisfactory performance metrics in the tests (accuracy = 0.879, sensitivity = 0.827, precision = 0.949, and F1-score = 0.887). These outcomes indicate that the network achieved a large number of correct answers and accurately identified patterns of early carious lesions on teeth with truly caries-affected occlusal surfaces (sensitivity), as well as patterns of health on truly healthy teeth (precision). Notably, only two studies assessed the effectiveness of CNNs in detecting early carious lesions on dental photographs and their findings were similar to ours [37, 38]. Kuhnisch et al. utilized 1,100 intraoral photographs of canines, premolars, and molars, previously assessed by the ICDAS, for MobileNetV2 learning, having obtained 85.2% of accuracy in the detection of early noncavitated carious lesions [37]. Considering the observed high efficiency of CNN in detecting early caries lesions, Li et al. conducted a study using 712 intraoral photographs and they found 95.8% sensitivity and 99% precision, but they did not employ an index for visual examination standardization [38]. Among the available systems, the ICDAS, used in the present study, proves to be more efficient in the detection of early carious lesions because it includes several clinical aspects, from the very first signs, and also because its diagnostic accuracy is akin to that of the histological analysis for detection of lesions on occlusal surfaces [22, 23]. Previous studies that used dental images for the detection of dental carious lesions have not described the different stages of carious lesions [36, 39–42] and they have not assessed the effectiveness of the network in the distinction between early carious lesions and opacities caused by developmental defects of enamel [35].

Even though the detection of dental caries has been extensively investigated for the development of CNNs, most studies have used radiographs as dental images for network learning [25, 51]. Nevertheless, these additional exams are not efficient in the detection of early carious lesions on occlusal surfaces because of incipient mineral loss by such lesions and overlap of buccal and lingual cusp enamel over the region of occlusal fissures [25, 37, 52]. On the other hand, photographs, used as input data for the network, have excellent cost efficiency for clinicians because they are a type of image that most frequently resembles those observed during visual examination; they are easily obtained, as they can be taken with the smartphone, digital camera, professional camera, or intraoral camera. Therefore, it should be underscored that this is one of the first studies to use dental photographs in deep learning by a CNN for the detection of early dental carious lesions. Furthermore, the utilization of a professional camera enables the capture of high-quality occlusal photographs, facilitating the visualization of milder lesions.

While the images used in this study are of extracted teeth, all photographs were pre-processed, providing similar images to those obtained by authors who conducted deep learning based on intraoral photographs, with tooth cropping and noise elimination, allowing the algorithm to focus specifically on the region of interest [35–42]. Moreover, the photographs were taken at different angles and distances from the tooth to simulate real scenarios of intraoral images capture, providing a larger number of images and different angles of the lesions to be trained by VGG-19.

The literature strongly suggests that the promising potential of automated methods in dentistry for diagnosing dental caries [26, 34, 39, 42]. To the best of our knowledge, however, none of the studies has assessed the efficiency of these algorithms as a diagnostic aid, as they have not evaluated examiners’ performance before and after the network results. Thus, this is the first study that evaluated such analysis, and it indicates that the CNN is highly effective in diagnosing early carious lesions on occlusal photographs even considering examiners with different levels of clinical experience. It is important to emphasize that the accuracy of examiners was still lower than that obtained from the algorithm in all test phases.

The present study has some limitations, considering that the extracted teeth used for image acquisition had been subjected to disinfection and storage, which might have altered the clinical appearance of the teeth, as compared to when they were still in the mouth. In addition, those teeth with extrinsic non-carious stains or subjected to restorative or endodontic procedures were excluded from the study. Therefore, the high effectiveness of VGG-19 observed herein should be viewed as a preliminary finding and to stimulate new research using intraoral photographs of teeth with different clinical aspects, enhancing the applicability of deep-learning networks in clinical dental practice.

VGG-19 deep-learning network demonstrated high efficiency in detecting healthy teeth and early carious lesions based on occlusal photographs of extracted teeth. It also proved to be a diagnostic aid in early detection of dental caries.

Author Contribution

Portella PD: Conceptualization, Methodology, Validation, Formal Analysis, Investigation, Resources, Data Curation, Writing - Original Draft. de Oliveira LF: Conceptualization, Methodology, Software, Formal Analysis, Writing - Review & Editing, Supervision. Ferreira MFC: Software, Formal Analysis. Dias BC: Investigation. de Souza JF: Investigation, Writing - Review & Editing. Assunção LRDS: Term, Conceptualization, Methodology, Validation, Formal Analysis, Investigation, Resources, Writing - Review & Editing, Project administration.

Ethics Approval and Consent to Participate

The study protocol was approved by the Human Research Ethics Committee of the Division of Health Sciences of the Universidade Federal do Paraná (CAAE 25001219.5.0000.0102). Test phase evaluators received an electronically informed consent form.

Funding Ethics Approval and Consent to Participate

This study was partially funded by the CAPES – Brazilian Federal Agency for Support and Evaluation of Graduate Education – Finance Code 001.

Conflict of Interests

The authors declare no conflicts of interest.

GBD 2017 Disease and Injury Incidence and Prevalence Collaborators. Global, regional, and national incidence, prevalence, and years lived with disability for 354 diseases and injuries for 195 countries and territories, 1990–2017: a systematic analysis for the Global Burden of Disease Study 2017. Lancet (2018) 10;392(10159):1789–1858. doi: 10.1016/S0140-6736(18)32279-7
Peres MA, Macpherson LMD, Weyant RJ, Daly B, Venturelli R, Mathur MR, Listl S, Celeste RK, Guarnizo-Herreño CC, Kearns C, Benzian H, Allison P, Watt RG. Oral diseases: a global public health challenge. Lancet (2019) 20;394(10194):249–260. doi: 10.1016/S0140-6736(19)31146-8.
Lima JEO. Cárie dentária: um novo conceito. Rev Dent Press Ortodon Ortop Facial (2007) 12(6). doi: 10.1590/S1415-54192007000600012
García-Pérez Á, Irigoyen-Camacho ME, Borges-Yáñez SA, Zepeda-Zepeda MA, Bolona-Gallardo I, Maupomé G. Impact of caries and dental fluorosis on oral health-related quality of life: a cross-sectional study in schoolchildren receiving water naturally fluoridated at above-optimal levels. Clin Oral Investig (2017) 21(9):2771–2780. doi: 10.1007/s00784-017-2079-1
Lacerda JT de, Castilho EA de, Calvo MCM, Freitas SFT de. Saúde bucal e o desempenho diário de adultos em Chapecó, Santa Catarina, Brasil. Cad Saúde Pública (2008) 24(8). doi: 10.1590/S0102-311X2008000800013
Mota-Veloso I, Soares ME, Alencar BM, Marques LS, Ramos-Jorge ML, Ramos-Jorge J. Impact of untreated dental caries and its clinical consequences on the oral health-related quality of life of schoolchildren aged 8–10 years. Qual Life Res (2016) 25(1):193–9. doi: 10.1007/s11136-015-1059-7
Onoriobe U, Rozier RG, Cantrell J, King RS. Effects of enamel fluorosis and dental caries on quality of life. J Dent Res (2014) 93(10):972–9. doi: 10.1177/0022034514548705
Ramos-Jorge J, Pordeus IA, Ramos-Jorge ML, Marques LS, Paiva SM. Impact of untreated dental caries on quality of life of preschool children: different stages and activity. Community Dent Oral Epidemiol (2014) 42(4):311–22. doi: 10.1111/cdoe.12086
Scarpelli AC, Paiva SM, Viegas CM, Carvalho AC, Ferreira FM, Pordeus IA. Oral health-related quality of life among Brazilian preschool children. Community Dent Oral Epidemiol (2013) 41(4):336–44. doi: 10.1111/cdoe.12022
Sheiham A, Steele JG, Marcenes W, Tsakos G, Finch S, Walls AW. Prevalence of impacts of dental and oral disorders and their effects on eating among older people; a national survey in Great Britain. Community Dent Oral Epidemiol (2001) 29(3):195–203. doi: 10.1034/j.1600-0528.2001
Boeira GF, Correa MB, Peres KG, Peres MA, Santos IS, Matijasevich A, Barros AJ, Demarco FF. Caries is the main cause for dental pain in childhood: findings from a birth cohort. Caries Res (2012) 46(5):488–95. doi: 10.1159/000339491
Brasil. SB Brasil 2010: Pesquisa Nacional de Saúde Bucal: resultados principais. Ministério da Saúde, 2012. bvsms.saude.gov.br/bvs/publicacoes/pesquisa_nacional_saude_bucal.pdf (accessed 1 Feb 2023).
World Health Organization (WHO). Oral health surveys: basics methods. 5th ed. Geneva: Word Health Organization, 2013.
Pitts NB, Baez RJ, Diaz-Guillory C, Donly KJ, Alberto Feldens C, McGrath C, Phantumvanit P, Seow WK, Sharkov N, Songpaisan Y, Tinanoff N, Twetman S. Early Childhood Caries: IAPD Bangkok Declaration. J Dent Child (Chic) (2019) 86(2):72.
Machiulskiene V, Campus G, Carvalho JC, Dige I, Ekstrand KR, Jablonski-Momeni A, Maltz M, Manton DJ, Martignon S, Martinez-Mier EA, Pitts NB, Schulte AG, Splieth CH, Tenuta LMA, Ferreira Zandona A, Nyvad B. Terminology of Dental Caries and Dental Caries Management: Consensus Report of a Workshop Organized by ORCA and Cariology Research Group of IADR. Caries Res. (2020) 54(1):7–14. doi: 10.1159/000503309
Peters MC, McLean ME. Minimally invasive operative care. I. Minimal intervention and concepts for minimally invasive cavity preparations. J Adhes Dent (2001) 3(1):7–16.
Ericson D, Kidd E, McComb D, Mjör I, Noack, MJ. Minimally Invasive Dentistry–concepts and techniques in cariology. Oral health & preventive dentistry (2003) 1(1), 59–72.
Marinho VA, Pereira GM. Revisão de literatura cárie: diagnóstico e plano de tratamento. Rev. Un. Alfenas (1998) 4: 27–37.
Clovis JB, Horowitz AM, Kleinman DV, Wang MQ, Massey M. Maryland dental hygienists' knowledge, opinions and practices regarding dental caries prevention and early detection. J Dent Hyg (2012) 86(4):292–305.
Manski MC, Parker ME. Early childhood caries: knowledge, attitudes, and practice behaviors of Maryland dental hygienists. J Dent Hyg (2010) 84(4):190–5.
Ismail AI, Sohn W, Tellez M, Amaya A, Sen A, Hasson H, Pitts NB. The International Caries Detection and Assessment System (ICDAS): an integrated system for measuring dental caries. Community Dent Oral Epidemiol (2007) 35(3):170–8. doi: 10.1111/j.1600-0528.2007.00347.x
Alaa E, Messiry He. Comparative Evaluation of ICDAS, WHO and Histological Examination in Detection of Occlusal Carious Lesions. Braz Dent Sci (2020) 23(3).
Carvalho RN, Letieri ADS, Vieira TI, Santos TMPD, Lopes RT, Neves AA, Pomarico L. Accuracy of visual and image-based ICDAS criteria compared with a micro-CT gold standard for caries detection on occlusal surfaces. Braz Oral Res (2018) 10;32:e60. doi: 10.1590/1807-3107bor-2018.vol32.0060
Gomez, J. Detection and diagnosis of the early caries lesion. BMC Oral Health (2015) 15(1). doi: 10.1186/1472-6831-15-S1-S3
Foros P, Oikonomou E, Koletsi D, Rahiotis C. Detection Methods for Early Caries Diagnosis: A Systematic Review and Meta-Analysis. Caries Res (2021) 55(4):247–259.
Hwang JJ, Jung YH, Cho BH, Heo MS. An overview of deep learning in the field of dentistry. Imaging Sci Dent (2019) 49(1):1–7. doi: 10.5624/isd.2019.49.1.1
Litjens G, Kooi T, Bejnordi BE, Setio AAA, Ciompi F, Ghafoorian M, van der Laak JAWM, van Ginneken B, Sánchez CI. A survey on deep learning in medical image analysis. Med Image Anal (2017) 42:60–88. doi: 10.1016/j.media.2017.07.005
Schwendicke F, Samek W, Krois J. Artificial Intelligence in Dentistry: Chances and Challenges. J Dent Res (2020) 99(7):769–774. doi: 10.1177/0022034520915714
Shan T, Tay FR, Gu L. Application of Artificial Intelligence in Dentistry. J Dent Res (2021) 100(3):232–244. doi: 10.1177/0022034520969115
Schmidhuber J. Deep learning in neural networks: An overview. Neural networks: the official journal of the International Neural Network Society (2015) 61:85–117.
Karimian N, Salehi HS, Mahdian M, Alnajjar, H, Tadinada A. Deep learning classifier with optical coherence tomography images for early dental caries detection. Proc. SPIE 10473, Lasers in Dentistry XXIV (2018) e: 1047304. doi: 10.1117/12.2291088
Lee JH, Kim DH, Jeong SN, Choi SH. Detection and diagnosis of dental caries using a deep learning-based convolutional neural network algorithm. J Dent (2018) 77:106–111. doi: 10.1016/j.jdent.2018.07.015
Casalegno F, Newton T, Daher R, Abdelaziz M, Lodi-Rizzini A, Schürmann F, Krejci I, Markram H. Caries Detection with Near-Infrared Transillumination Using Deep Learning. J Dent Res (2019) 98(11):1227–1233. doi: 10.1177/0022034519871884
Schwendicke F, Golla T, Dreher M, Krois J. Convolutional neural networks for dental image diagnostics: A scoping review. J Dent (2019) 91:103226. doi: 10.1016/j.jdent.2019.103226
Askar H, Krois J, Rohrer C, Mertens S, Elhennawy K, Ottolenghi L, Mazur M, Paris S, Schwendicke F. Detecting white spot lesions on dental photography using deep learning: A pilot study. J Dent (2021) 107:103615. doi: 10.1016/j.jdent.2021.103615
Ding B, Zhang Z, Liang Y, Wang W, Hao S, Meng Z, Guan L, Hu Y, Guo B, Zhao R, Lv Y. Detection of dental caries in oral photographs taken by mobile phones based on the YOLOv3 algorithm. Ann Transl Med (2021) 9(21):1622. doi: 10.21037/atm-21-4805
Kühnisch J, Meyer O, Hesenius M, Hickel R, Gruhn V. Caries Detection on Intraoral Images Using Artificial Intelligence. Journal of Dental Research (2022) 101(2):158–165. doi:10.1177/00220345211032524
Li RZ, Zhu JX, Wang YY, Zhao SY, Peng CF, Zhou Q, Sun RQ, Hao AM, Li S, Wang Y, Xia B. [Development of a deep learning based prototype artificial intelligence system for the detection of dental caries in children]. Zhonghua Kou Qiang Yi Xue Za Zhi (2021) 9;56(12):1253–1260. doi: 10.3760/cma.j.cn112144-20210712-00323
Park EY, Cho H, Kang S, Jeong S, Kim EK. Caries detection with tooth surface segmentation on intraoral photographic images using deep learning. BMC Oral Health (2022) 7;22(1):573. doi: 10.1186/s12903-022-02589-1
Rashid U, Javid A, Khan AR, Liu L, Ahmed A, Khalid O, Saleem K, Meraj S, Iqbal U, Nawaz R. A hybrid mask RCNN-based tool to localize dental cavities from real-time mixed photographic images. PeerJ Computer Science (2022) 8:e888 https://doi.org/10.7717/peerj-cs.888
Zang XY, Qiao B, Meng FH, Jin NH, Hu SX, Li LB, Xing LJ, Chen F, Wang Y, Zhang HZ. [A deep learning segmentation model for detecting caries in molar teeth]. Zhonghua Yi Xue Za Zhi (2022) 30;102(32):2538–2540. doi: 10.3760/cma.j.cn112137-20220422-008957
Zhang X, Liang Y, Li W, Liu C, Gu D, Sun W, Miao L. Development and evaluation of deep learning for screening dental caries from oral photographs. Oral Dis (2022) 28(1):173–181. doi: 10.1111/odi.13735
Bossuyt PM, Reitsma JB, Bruns DE, Gatsonis CA, Glasziou PP, Irwig L, Lijmer JG, Moher D, Rennie D, de Vet HC, Kressel HY, Rifai N, Golub RM, Altman DG, Hooft L, Korevaar DA, Cohen JF. STARD 2015: an updated list of essential items for reporting diagnostic accuracy studies, Bmj 351 (2015) h5527.
Mongan Moy JL, Kahn CE. Checklist for artificial intelligence in medical imaging (CLAIM): a Guide for authors and reviewers, Radiol. Artif. Intell. 2 (2) (2020), e200029.
Landis J. R., Koch G. G. The measurement of observer agreement for categorical data. Biometrics (1977) 33(1):159–174.
YU JSTS. Label image regions. 2013.
Simonyan K, Zisserman A. Very Deep Convolutional Networks for Large-Scale Image Recognition, International Conference on Representation, Vancouver, 2015.
Russakovsky O, Deng J, Su H, Krause J, Satheesh S, Ma S, Huang Z, Karpathy A, Khosla A, Bernstein M, Berg AC, Fei-Fei L. ImageNet Large Scale Visual Recognition Challenge. International Journal of Computer Vision, 2015.
Deng J, Dong W, Socher R, Li LJ, Li K, Fei-Fei L. ImageNet: A Large-Scale Hierarchical Image Database. IEEE Computer Vision and Pattern Recognition (CVPR), 2009.
Lee JM, Shin SC, Cho JW, Choi YH, Moon YM, Jung SJ, Kwon JH. International Journal of ... The Evaluation for Oral Examination by Using ofIntra-Oral Camera International journal of clinical preventive dentistry (2014) 10(2):113–120.
Mohammad-Rahimi H, Motamedian SR, Rohban MH, Krois J, Uribe SE, Mahmoudinia E, Rokhshad R, Nadimi M, Schwendicke F. Deep learning for caries detection: A systematic review. J Dent (2022) 122:104115. doi: 10.1016/j.jdent.2022.104115
Dayo AF, Wolff MS, Syed AZ, Mupparapu M. Radiology of Dental Caries. Dental clinics of North America (2021) 65(3):427–445.

No competing interests reported.

Download PDF

Journal Publication

published 31 Oct, 2023

Read the published version in Clinical Oral Investigations →

Editorial decision: Major revision
03 Aug, 2023
Reviews received at journal
29 Jul, 2023
Reviewers agreed at journal
19 Jul, 2023
Reviewers agreed at journal
13 Jul, 2023
Reviewers invited by journal
13 Jul, 2023
Editor assigned by journal
05 Jul, 2023
Submission checks completed at journal
05 Jul, 2023
First submitted to journal
29 Jun, 2023

You are reading this latest preprint version

Deep-learning-based automated detection and assistance in the diagnosis of early dental carious lesions

Status:

Journal Publication

Version 1

Abstract

Objective

Materials and Methods

Results

Conclusions

Clinical relevance:

Figures

INTRODUCTION

MATERIALS AND METHODS

Ethical aspects

Sampling

Clinical examination

Image databases

Deep-learning model

VGG-19 training and test

RESULTS

Database characteristics

VGG-19 performance metrics

Diagnostic aid

DISCUSSION

CONCLUSION

Declarations

References

Additional Declarations

Status:

Journal Publication

Version 1