Inspection of conditions such as color changes, contour changes, gingival hyperplasia in the gingiva is one of the most important stages in periodontal clinical examination and disease determination [7, 36]. In addition, conditions such as deep frenulum attachments and insufficient keratinized gingiva areas should be recorded by physicians during examinations. Because these situations can disrupt the continuity of oral hygiene and sometimes cause aesthetic problems; there may be a need for surgical periodontal treatment in the relevant field in the future [7, 36]. Periodontists are specialists in the field of periodontology, a specialized field of dentistry, who carry out periodontal treatment of patients by making detailed periodontal evaluation and perform periodontal surgical treatments when necessary [37]. General dentists and dentists specialized in other fields may be inadequate in the periodontal examination and oral-mucosal status evaluation of the patients in some cases, or they may have to refer the patient to the periodontist in specialized cases [38]. It is recommended to work in coordination with a periodontist in patient planning and follow-up, especially in specialties such as orthodontics [39, 40]. However, it is not always possible to work in coordination with a periodontology specialist and to reach a specialist physician. In addition, due to the intensity, fatigue and lack of experience of the physician, some cases may be overlooked by the periodontologist. This situation brings to mind the usability of computer-aided systems that can provide decision support mechanism to physicians for diagnostic purposes.
Based on this information, in the current study, it was aimed to develop an AI system that automatically evaluates periodontal status from intraoral photographs, which are widely used for patient follow-up and recording in all areas of dentistry, and to measure the success of this system. In this study, the success of AI supported systems in the detection of periodontal disease findings such as erythematous-inflamed gingiva and enlarged gingiva and anatomical structures such as the frenulum, which plays an important role in mucogingival evaluation, was evaluated.
As in many fields of medicine, the usability of AI systems in image processing and interpretation in dentistry, especially in areas such as radiology and pathology, has been demonstrated, and it has been confirmed by academic studies that they can be a decision-support mechanism for physicians [16–19]. When the AI-based periodontology studies are examined, it is seen that these systems are very successful in the interpretation of intraoral and extraoral dental radiographs; also, it is seen that information such as the presence of bone loss, the presence of periodontal disease and the severity of the disease can be determined automatically with AI [23, 41, 42].
In the literature, detection of lesions such as squamous cell carcinoma [26], lichen planus [43] using AI systems in intraoral photographs; detection of dental applications such as dental prostheses, restorations and fissure sealants [29, 44, 45]; in addition, there are many studies on the detection of conditions such as dental caries [28, 46], white spot [18], and anomalies such as microdontia, rotation, and supernumerary [47]. All of these studies about AI, which has attracted great interest in dentistry in recent years, support the usability of these systems for intraoral photographs and dental cameras in the dental field in the coming years.
In one of these studies, Fu et al. (2020) had tried to determine oral squamous cell carcinoma with CNN using 1469 intraoral photographs and compared the results of 21 observers (expert, medical student, non-medical student) with the results of AI in their study [26]. As a result of this study, they had reported that AI systems have a performance comparable to the specialist physician in determining the relevant pathology, and that the system performs significantly better than an average medical student [26]. Similarly, Keser et al. (2021), in their study, had tried to determine the oral lichen planus lesions related using inception v3 architecture (an AI system) on 65 intraoral photographs of healthy induviduals and 72 patients with oral lichen planus [43]. They also had reported that the AI system used in this study was 100% successful in detecting the presence/absence of the lesion, that is, in its classification [43].
Zhang et al. (2020) had carried out a study on the automatic detection of dental caries by CNN-based AI systems on a large dataset containing 3932 photographs [28]. As a result of this study, they had reported that the system gave very successful results in detecting the related pathology and that the use of these systems in caries screening in crowded populations could be beneficial [28]. Kühnisch, et al. (2022) also had carried out a similar study and reported that CNN-based AI systems can be used in caries detection, but these systems need to be improved [46]. Although different situations have been evaluated with AI systems, our study shows that AI systems can be successfully used in the future in determining dental status, tooth numbering and anatomical formation from intraoral photographs similarly.
Takahashi et al. (2021) had found that metallic colored restorations were recognized by the ai-systems with a higher accuracy rate, as a result of their study on prosthesis and restoration detection using AI systems (This study was carried out on 1904 dental photographs) [29]. In another study, Engels et al. (2022) had reported that AI systems showed success in the range of 92.9–99.2% in the determination of different restorations such as non-restorative tooth, composite restoration, cement restoration, amalgam restoration, gold restoration and ceramic restoration [44]. On the other hand, Schlickenrieder et al. (2021) had reported the usability of CNN systems in the determination of fissure sealant in intraoral photographs. Even if the related studies are not directly related to the field of periodontology, it supports that all kinds of pathology, restoration and dental conditions can be detected with AI systems [45].
When the literature is examined, it can be seen that these systems can be used in a wide variety of pathology detections. Askar et al. (2021), in their study using 434 photographs, had reported that CNN systems showed satisfactory results in the determination of cases such as White spot and fluorosis, and that new studies should be carried out on large data sets for more successful results [18]. Rogodos et al. (2022) had used 38486 intraoral photographs to evaluate 10 dental anomalies including rare anomalies (such as mammalons, hypoplasia, Microdontia) by AI systems and obtained successful results [47].
It is known that the success of the system increases as the number of data increases in AI systems. Therefore, in the current study, the number of data was tried to be kept as wide as possible. As can be seen, although many conditions such as caries, restoration, prosthesis, and anomalies have been evaluated with AI systems in the literature, tooth numbering has not been performed in any of these studies. In the present study, the system's ability to identify teeth and perform tooth numbering was also studied. Thus, it is aimed to analyze the related pathology, dental disease and which tooth is associated with the condition. As far as we know, the present study is the first study in which tooth numbering is done from photographs using AI systems. This parameter is very important for dental evaluations. Because, not only in periodontology, but also in all AI studies to be developed in the dental field, it allows the automatic determination of which tooth is associated with the relevant pathology, restoration or dental condition. In summary, It can enable the localization of related pathologies and conditions to be determined and subsequently converted into a written report.
The studies most similar to the aim of the present study were studies of Alalharith et al. (2020), You et al. (2020) Xu et al. (2022) and Li et al. (2021) [27, 30, 48, 49]. Because in these studies, AI systems that will help the physician in periodontal evaluation such as the detection of gingivitis findings and the determination of dental plaque, as well as inform the patient, had been studied. Alalharith et al. (2020) had tried to detect tooth detection and gingival inflammation using CNN systems on 134 photographs [48]. This study also had evaluated inflammatory and erythematous site detection, similar to the current study. But Alalharith et al. (2020) had used the object detection method in their studies [48], while the segmentation method, which is a more ideal method, was used in the current study. It can be said that the current study is more comprehensive and superior to this study in terms of it has larger data set, the use of segmentation method in evaluations, and the evaluation of different parameters. On the other hand, in the study carried out by Li et al. (2021), 3932 intraoral photographs obtained from 625 patients were used [30]. In this study, similar to the study of Alalharith et al. (2020), they had tried to determine the signs of gingivitis, soft appendages and calculus with the objective detection method [48]. Although similar results and system success were reported in the detection of inflammation with the results of the current study, the fact that the object detection method was used in this study can be considered as a disadvantage of the study.
You et al. (2020) had used the CNN method to detect dental plaque around primary teeth and reported that AI systems detected plaque at a higher accuracy rate than the dentist in their study on 886 photographs. In this study, the segmentation method was used and in this respect, it is similar to the current study [27]. However, the fact that only one parameter had been evaluated and studied only in primary tooth photographs can be considered as a disadvantage of this study.
Xu et al. (2022) had carried out a study using the Googlenet model, and in this study, they had aimed to analyze the plaques around the teeth with the object detection method [49]. Only 400 images were used in the study of Xu et al. (2022) [49]. In these two studies, they had worked on a critical parameter in terms of periodontology, namely the determination of plaque, and presented acceptable success rates [27, 49]. It should not be forgotten that plaque control is the basic principle in the treatment of periodontal disease, and this issue should be more comprehensively focused in future AI-based periodontology studies on the intraoral photographs.
Although the related study presented in the article had not been evaluated before in AI-based studies performed in photographs and had evaluated various parameters that may be periodontally important, it had some limitations. The first of these was the limited number of data and the fact that it was carried out only on intraoral facade photographs. Studies in which the type of photography is diversified by increasing the number of data will provide more meaningful results. In addition, the decisions of many auditors were not recorded and were not compared with the relevant CNN system. These can be considered as the limitations of the study. Had the study been conducted in this way, it would undoubtedly have presented more interpretable results. These issues should be taken into consideration in future studies, and more comprehensive studies should be carried out with photographs taken from different cameras and of different quality, in which more parameters are evaluated. It is obvious that AI systems, which gained importance in many fields, will take place in the diagnosis and treatment planning in dentistry in the coming years. Academic studies to be carried out will accelerate this acceleration.