search for




 

Development of AI Web Service for Quantification of Dental Plaque
Int J Clin Prev Dent 2024;20(1):27-32
Published online March 31, 2024;  https://doi.org/10.15236/ijcpd.2024.20.1.27
© 2024 International Journal of Clinical Preventive Dentistry.

Jae-Young Lee1,2, Ji-Na Lim2,3, Byung-Hee Han4, Soo-Hwang Seok2, Hyun-Jun Yoo5

1Department of Dental Hygiene, College of Health Science, Dankook University, Cheonan, 2Research Institute, THOMASTONE Co., Ltd, Cheonan, 3Department of Public Health Science, Graduate School, Dankook University, Cheonan, 4Research Institute, Kai-i Company, DaeJeon, 5Department of Preventive Dentistry, College of Dentistry, Dankook University, Cheonan, Korea
Correspondence to: Jae-Young Lee
E-mail: dentaljy@dankook.ac.kr
https://orcid.org/0000-0003-2394-5894
Received March 2, 2024; Revised March 8, 2024; Accepted March 8, 2024.
This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (http://creativecommons.org/licenses/by-nc/3.0) which permits unrestricted non-commercial use, distribution, and reproduction in any medium, provided the original work is properly cited.
Abstract
Dental plaque may be detected using a plaque disclosing solution. Upon application by a dental professional, the plaque disclosing solution enhances the visibility of dental plaque on the teeth and gingiva for clinical evaluation. This process takes approximately 10 minutes. The scores are subjective and based on professional judgment. Thus, the reliability of the score may be relatively low. Artificial intelligence (AI) and machine learning may improve the objectivity and reliability of dental plaque disclosing, To generate training data for machine learning, large amounts of oral data relating to the detection of dental plaque over the same area need to be collected. A system was developed to support data storage management, classification, and the creation of training data. The YOLOv7 model was adopted. This AI model demonstrated the best performance in real-time dental plaque detection. Moreover, the related app service facilitates the generation of real-time results from clinical data, and provides a web-based administrative server service that can be used to organize patient data. This study demonstrated that the developed model is effective for dental plaque evaluation.
Keywords : aritificial intelligence, dental plaque
Introduction

The application scope of Artificial Intelligence (AI) technology in the medical field has significantly expanded, and interventions using AI technology and big data have become capable of delivering high-precision, personalized medical care with greater accuracy and efficiency, overcoming many drawbacks of human performance [1]. Additionally, AI technology offers the significant advantage of supplementing scarce human resources [2] and promoting continuous self- management. These AI qualities are utilized for managing chronic diseases and correcting health behaviors that require ongoing self-management [3]. Moreover, AI has the capacity to consider a wide range of information about an individual, which is essential for perpetual, customized, patient-centered interventions. Importantly, one-size-fits-all approaches are no longer desirable and AI technology serves as an excellent tool for improving various forms of personalized health care interventions [4]. Most individuals struggle to maintain self- care in the absence of external health care interventions [5]. Strategies to assist with forming health behavior habits as part of continuous and effective self-management are warranted. The personalized capabilities of AI may be helpful for improving personalized dental care, especially in supporting effective health habit interventions [6].

Dental plaque, a major cause of the two primary oral diseases, dental caries and periodontal disease, is a community of microorganisms living in the oral cavity, forming a complex ecosystem on the surface of the teeth [7]. The management of dental plaque is important, not only for maintaining personal oral hygiene, but also contributes to the overall health and well-being of an individual. For example, dental plaque can influence the development and progression of heart and metabolic diseases [8,9]. Recent studies have highlighted the interconnection between oral health and overall health, underscoring the importance of managing dental plaque [10]. The coronavirus disease 2019 (COVID-19) pandemic has led to the suspension of community face-to-face oral health programs and education in public facilities such as kindergartens, daycare centers, and schools, with the provision of public oral health care services being partially suspended or limited [11]. As efforts to minimize direct human contact and reduce offline activities continue, the adoption of contactless or automated systems in various fields is increasing [12]. Despite the return to daily life and the normalization of health services in the post-corona era, it is difficult to develop oral health services as actively as before [13]. In this context, digital healthcare in a contactless form can be an important tool, and its significance in the field of oral care is increasingly highlighted [14]. Developing and implementing contactless healthcare services that assess health status and provide information is crucial [15]. This study aimed to develop and assess an image-based AI technology for quantifying dental plaque, thus addressing one of the most significant causes of oral diseases.

Materials and Methods

1. Development of AI web service for quantifying dental plaque

1) System overview

In this paper, we designed an AI app service to quantify dental plaque. Figure 1 shows the overall system structure. The proposed app service performs the function of quantifying and locating oral plaque stained with a dental stain (Dentichak, Thomastone Co., Ltd, Republic of Korea) that specifically stains dental plaque within the oral cavity. Dental plaque is detected and analyzed based on tooth images taken with a smartphone. The detected data confirm the location of dental plaque for analysis. Images taken are uploaded to the service app. The entire oral cavity is recognized. Information for each tooth range (e.g., maxilla-left/right, mandible-left/right, quadrant) is delivered to the service backend. The AI models an inference from the independently captured oral images, which is then used to optimize the model inference speed. Meanwhile, the functions related to the dental plaque monitoring service app contribute to the backend processes.

Figure 1. Overall system structure of wed/app service for diagnosis of dental plaque.

2. Development of AI model for quantifying dental plaque

1) Data collection

In this study, oral imaging image data for AI learning was collected from May 2022 to October 2023. Continuous collection was conducted according to the collection procedure, and the final data was collected as 5,000 cases (Figure 2).

Figure 2. Overall system structure of wed/app service for diagnosis of dental plaque.
2) Training data generation

An AI learning verification process was carried out on the dataset, wherein the existing AI image learning dataset was divided according to the 6:2:2 ratio to create training data. A web-based labeling tool was developed to facilitate usage across various computer environments, and this tool was employed to label dental plaque data extracted from the collected dataset. Both Bounding Box and Polygon areas were designated for marking, and labeling was conducted within the Canvas area using a class designation feature to identify teeth and oral plaque regions. After delineating the areas of interest and classes for the labeling process, adjusting positions, and revising the labels, the AI learning data was compiled through an Export function, thereby assembling the labeled data for training.

The dental biofilm detection AI model was developed through the labeling process for training, verification, and testing of both the tooth recognition (gingival tooth separation) model and the tooth detection (plaque detection) model. For the tooth recognition model, the gingival range for 12 positions on the frontal teeth was delineated. As for the tooth detection model, labeling was performed on both teeth and plaque areas (Figure 3).

Figure 3. Database structure of medical image management and labeling system for diagnosis of dental plaque.
3) Development and learning of AI model for quantifying dental plaque

YOLOv7 was finally selected as the tooth recognition model after performance verification, and Deeplav3 was finally selected as the tooth detection model, and each model has the most accurate and fastest detection ability through AI model performance verification. We verified the suitability of performance by comparing YOLOv4 and YOLOv7 as tooth detection models, and UNet and Deeplav3 as tooth detection models. We consulted dentists and dental hygienists at dental university hospitals to verify the accuracy of the inference results of teeth and dented plaque ranges using actual images. Verification of the two models was carried out simultaneously. The object naming process for the tooth recognition model was learned by specifying the tooth standards of 12 frontal photos of the mouth and the dental plaque detection model matched individual teeth and plaque areas that were deemed the same. After comparing performance, the optimal learning model was selected and advanced.

4) Evaluation of the AI model and agreement with dental professionals’ annotations

To validate the clinical relevance, a minimum of 1,000 photographic data were collected using a digital camera (3,216× 2,136 pixels, Canon EOS 60D, Japan), and an additional 100 photographs were taken using the cameras integrated into smart-phones (1,280×960 pixels, iPhone 14 pro [Apple], Gaxlaxy 23 Ultra [Samsung]). Both types of photographs depicted the dental plaque on the tooth surface stained with a plaque disclosing solution (plaque checker), and were analyzed and annotated for dental plaque by both dental professionals and the AI program. The agreement between the evaluations made by the AI model and the markings by professionals was confirmed, and to assess the consistency of manual diagnoses, the same photographs were evaluated by different dental profes-sionals. In another comparison, 100 photographs of primary teeth taken with an intraoral camera (1,280×960 pixels, TPC Ligang, Dongguan, China) were annotated to display the dental plaque areas evaluated by both the AI model and dental professionals. This approach improves diagnostic accuracy for each method based on photographs with lower resolution (fewer pixels) compared to images obtained with digital cameras.

Conclusion

This study described the entire development and machine training process for a dental plaque evaluation system. We provided details of all the steps, from uploading photographs of anterior teeth stained for dental plaque, to generating examination data and labelling data for AI learning, and analyzing dental plaque management through an AI model. The AI learning outcomes, utilizing YOLOv7, predict accuracy without the need for data preprocessing, enabling the analysis of normal and additional dental plaque content. However, it is anticipated that better results could be achieved by acquiring more dental plaque data to enhance the accuracy of the dental plaque AI diagnostic model. While dental plaque examination itself is straightforward in the clinical setting, the evaluation of dental plaque necessitates detailed analysis, consistent guidelines, and continuous follow-up management. Due to limited resources and time constraints in dental clinics, these procedures are often not performed. As a result, despite the necessity for adequate oral hygiene management following dental treatment, the dental professional focus remains solely on treatment provision. Consequently, the subsequent follow-up management is often neglected. The AI model described in this study may assist with simplifying this workflow process in the clinical setting.

Discussion

In the exploration of modern healthcare paradigms, the integration of Artificial Intelligence (AI) and digital platforms in delivering medical services signifies a pivotal shift towards more personalized and accessible treatment [16]. The model proposed in this study for quantifying and managing dental plaque is evidence of such advancements, showcasing a framework designed to navigate the complexities of healthcare environment while emphasizing the ultra-personalized centrality of individualized dental care and diagnosis [17]. Utilizing the capabilities of on-device AI, this model aims to extend effective dental health services beyond traditional face-to-face consultations to a digital format, enabling patients to receive timely and customized education and advice for managing dental plaque, a critical component in oral health maintenance [18]. The adjunctive use of AI technology in dental plaque evaluation could be likened to AI-assisted medical diagnosis technology. AI technology serves a supplemental role, rather than as a replacement for professional clinical judgements [19].

Despite its promising outlook, this model’s practical application is challenged by the need for robust continuous service management to fully realize its potential. A comprehensive understanding of oral health and its close relationship with systemic diseases is necessary [20]. This technology must also reflect the demands for patient-wide health data, such as MyData for medical diagnosis. Furthermore, this technology must comply with current assessment criteria [21]. Additionally, securing devices for capturing images of plaque on the teeth’s external and lingual surfaces could allow for broader AI learning and judgment capabilities. This technology could be integrated and utilized across various fields, including post-implant care and tartar removal procedures. Developing an agile management system capable of integrating new research findings, updating educational content, and adapting to regulatory changes without interrupting service delivery to the end-user is essential [22].

Another critical aspect is enhancing user experience. The effectiveness of digital health interventions heavily depends on their ability to engage users effectively. In the context of the dental plaque management education program, this entails creating an interactive and user-friendly interface that promotes ongoing engagement and facilitates the easy understanding of complex health information [23]. Personalization features adjusting advice and recommendations based on the user’s specific health data and preferences could significantly enhance the program’s relevance and impact. However, achieving such a level of personalization requires sophisticated algorithms and a deep understanding of user behaviors. The incorporation of human-centered design principles into model development is critical. This technology may also be utilized as a tool to assist in raising individual awareness of of daily oral health status, rather than as a diagnostic tool.

The integration and security of medical data represent another crucial focus area. The personalized nature of the proposed model heavily depends on the accurate and secure handling of sensitive health data [24]. Thus, ensuring the integrity and confidentiality of patient information is paramount. The model must also seamlessly integrate with existing healthcare IT systems to facilitate the efficient exchange and analysis of health data, thereby enhancing continuity of care. Addressing these issues requires a multidisciplinary approach, combining insights from healthcare professionals, AI and machine learning experts, data privacy specialists, and user experience designers. By fostering collaboration across these diverse fields, the model can be refined to meet the high standards of reliability, usability, and security required for successful implementation.

In conclusion, the dental plaque management education program model has the potential to revolutionize how dental care, as well as education and daily dental management, is provided and experienced. By attempting to identify and solve oral hygiene problems confirmed by individuals or professionals, this model can pave the way for a new era of dentistry that is more accessible, personalized, and effective. This model can break the stagnation in oral care, address oral inequality, and ease accessibility, allowing for applications across various fields.

Acknowledgements

This work was supported by Institute of Information & communications Technology Planning & Evaluation (IITP) grant funded by the Korea government (MSIT) (No. RS-2023- 00232355, Personalized oral care services based on oral care prescription data and dental photos).

Conflict of Interest

The plaque disclosing solution used in this study was provided free of charge by THOMASTONE Company (Cheonan, Republic of Korea) and was utilized as the material for research.

References
  1. Lee SB, Song J, Park A: A trend of artificial intelligence in healthcare. J Korea Contents Assoc 20: 448-56, 2020.
  2. McShane M, Beale S, Nirenburg S, Jarrell B, Fantry G: Incon-sistency as a diagnostic tool in a society of intelligent agents. Artif Intell Med 55: 137-48, 2012.
    Pubmed CrossRef
  3. Kwon DC, Jeong EK: National health and nutrition survey - 2019 national health statistics, Seoul Food Life Support Center, Seoul, 2019.
  4. Garden R: Expanding clinical empathy: an activist perspective. J Gen Intern Med 24: 122-5, 2009.
    Pubmed KoreaMed CrossRef
  5. Zhou M, Fukuoka Y, Mintz Y, Goldberg K, Kaminsky P, Flowers E, et al.: Evaluating machine learning-based automated personalized daily step goals delivered through a mobile phone app: randomized controlled trial. JMIR Mhealth Uhealth 6: e28, 2018.
    Pubmed KoreaMed CrossRef
  6. Shrivastava SR, Shrivastava PS, Ramasamy J: Role of self-care in management of diabetes mellitus. J Diabetes Metab Disord 12: 14, 2013.
    Pubmed KoreaMed CrossRef
  7. Marsh PD, Bradshaw DJ: Dental plaque as a biofilm. J Ind Microbiol 15: 169-75, 1995.
    Pubmed CrossRef
  8. Herzberg MC, Weyer MW: Dental plaque, platelets, and cardiovascular diseases. Ann Periodontol 3: 151-60, 1998.
    Pubmed CrossRef
  9. Saito T, Shimazaki Y: Metabolic disorders related to obesity and periodontal disease. Periodontol 2000 43: 254-66, 2007.
    Pubmed CrossRef
  10. Genco RJ, Sanz M: Clinical and public health implications of periodontal and systemic diseases: an overview. Periodontol 2000 83: 7-13, 2020.
    Pubmed CrossRef
  11. Kim MH, Hong EJ, Kwon YJ, Hwang YS: Impact of the COVID-19 pandemic on oral healthcare among children and teachers in kindergarten and daycare centers. J Korean Soc Dent Hyg 22: 487-96, 2022.
    CrossRef
  12. Song SS, Kwon KH, Kim HY: A Study on the Non-Face-to-Face Medical Communication Direction through Exploring of Tele-medicine Services Status and Abroad Case Studies. J Pract Res Advert Public Relations 14: 60-88, 2021.
  13. Kim TW, Lee JM: An evaluation of public responses taken in the area of income security after the Covid-19 pandemic. Health Welf Policy Forum 318: 22-33, 2023.
  14. Alonso SG, de la Torre-Díez I, Hamrioui S, López-Coronado M, Barreno DC, Nozaleda LM, et al.: Data mining algorithms and techniques in mental health: a systematic review. J Med Syst 42: 161, 2018.
    Pubmed CrossRef
  15. Cresswell K, Callaghan M, Khan S, Sheikh Z, Mozaffar H, Sheikh A: Investigating the use of data-driven artificial intelligence in computerised decision support systems for health and social care: a systematic review. Health Informatics J 26: 2138-47, 2020.
    Pubmed CrossRef
  16. Ahmed Z, Mohamed K, Zeeshan S, Dong X: Artificial intelligence with multi-functional machine learning platform development for better healthcare and precision medicine. Database 2020: baaa010, 2020.
    Pubmed KoreaMed CrossRef
  17. Shah V: Next-generation artificial intelligence for personalized medicine: challenges and innovations. Int J Comput Sci Tech 2: 1-15, 2018.
  18. Tjoa E, Guan C: A survey on explainable artificial intelligence (XAI): toward medical XAI. IEEE Trans Neural Netw Learn Syst 32: 4793-813, 2021.
    Pubmed CrossRef
  19. Jin W, Li X, Fatehi M, Hamarneh G: Guidelines and evaluation of clinical explainable AI in medical image analysis. Med Image Anal 84: 102684, 2023.
    Pubmed CrossRef
  20. Li X, Kolltveit KM, Tronstad L, Olsen I: Systemic diseases caused by oral infection. Clin Microbiol Rev 13: 547-58, 2000.
    Pubmed KoreaMed CrossRef
  21. Kim SJ, Ryu SH, Lim JN, Kwon HJ, Wang ZL, Lee JY, et al.: Constructing a non-face-to-face oral health care service model using my data: proposal. J Kor Acad Oral Health 47: 228-34, 2023.
    CrossRef
  22. Kim JS, Oh SH: Issue Analysis on the deregulation of telemedicine in the digital convergence era. J Digit Converg 18: 445-57, 2020.
  23. Akmal Muhamat N, Hasan R, Saddki N, Mohd Arshad MR, Ahmad M: Development and usability testing of mobile application on diet and oral health. PLoS One 16: e0257035, 2021.
    Pubmed KoreaMed CrossRef
  24. Fernández-Alemán JL, Señor IC, Lozoya PÁ, Toval A: Security and privacy in electronic health records: a systematic literature review. J Biomed Inform 46: 541-62, 2013.
    CrossRef


March 2024, 20 (1)