Intelligent Detection of Residual Carious Tissue During Caries Removal Using an Improved YOLOv8 Model

Intelligent Detection of Residual Carious Tissue During Caries Removal Using an Improved YOLOv8 Model | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Intelligent Detection of Residual Carious Tissue During Caries Removal Using an Improved YOLOv8 Model Ke Wang, Xulan Li, Jing Lai, Shanshan Guo, Yanming Chen, Fenfen Zhang, and 3 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-6933790/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Objective : This study aimed to develop a prototype of an AI-based recognition system for identifying residual carious tissue during caries removal by improving the YOLOv8 model. The goal was to provide preliminary theoretical and technical support for real-time AI-assisted guidance and localization of residual caries. Methods : A total of 100 extracted teeth with carious lesions were collected, numbered, and randomly assigned to training, validation, and test sets at a ratio of 7:1:2. During the extracoronal caries removal process for each tooth, 4 to 7 consecutive images were captured until all carious tissue was removed. Carious regions in the training images were manually annotated and used to train the improved YOLOv8 model. The model's predictions on the test set were compared with those made by standardized, experienced dental professionals. Performance was evaluated using accuracy, sensitivity, and specificity. Results : A total of 666 images were collected, including 566 images with carious lesions and 100 without. Compared with expert annotations, the improved YOLOv8 model achieved an accuracy of 87.9%, sensitivity of 88.4%, and specificity of 85% in identifying the presence of residual caries. For the localization of specific residual carious sites, the model achieved an accuracy of 92.8%, sensitivity of 93.2%, and specificity of 91.9%. Conclusion : The improved YOLOv8 model demonstrated strong capability in detecting and localizing residual carious tissue across consecutive images with high stability, indicating promising potential for clinical application. YOLOv8 Deep learning Dental caries Residual carious tissue Figures Figure 1 Introduction Dental caries is one of the most prevalent oral diseases worldwide [1], significantly impacting individuals’ oral health and overall quality of life. In the treatment of dental caries, the complete removal of decayed tissue is directly related to both therapeutic efficacy and long-term prognosis [2]. Currently, most clinicians rely primarily on direct visual inspection and tactile examination using a dental probe to assess the color and texture of residual tissue [3]. This approach is inherently subjective and heavily dependent on the operator’s clinical experience, lacking standardization and objectivity. With the continuous advancement of artificial intelligence (AI), the digitalization and automation of diagnostic and therapeutic procedures have become a major trend in modern medicine [4,5]. At present, AI technologies—particularly deep learning—can simulate the human brain’s ability to process medical images through multilayer neural networks [6–8]. Meanwhile, computer vision enables machines to understand and interpret visual information in images and videos [9,10]. For example, image segmentation techniques can isolate teeth and gingiva from the background to facilitate accurate health assessment [11], while object detection algorithms can identify specific targets within images. These automated processes not only improve the efficiency of image analysis but also reduce human error, enhancing the reliability of clinical decision-making. In recent years, AI has made rapid progress in the field of dentistry, demonstrating promising applications in diagnostic assistance, treatment planning, and beyond [12–15]. AI-assisted precision procedures have also begun to find practical use in clinical settings [16–19]. As a result, the development of an AI system capable of real-time guidance and localization of residual carious tissue during caries removal holds great potential. Such a system could improve the quality of operative procedures and lay the groundwork for the future realization of automated caries removal, marking a significant advancement in the application of AI in operative dentistry and endodontics. To achieve this goal, the primary technical challenge is enabling rapid and accurate identification and localization of residual carious tissue during dynamic operative procedures. This would allow the task of intraoperative assessment, traditionally carried out by clinicians, to be assisted or even replaced by an AI system. YOLOv8(You Only Look Once, Version 8), as a new-generation object detection algorithm, offers faster computational speed and higher detection accuracy[20,21]. Its application in identifying residual carious tissue during tooth preparation holds promise for enhancing the intelligence level of dental diagnosis and treatment, enabling real-time assistance in localization and assessment. This study explores the feasibility of using an improved YOLOv8 model to detect and locate residual carious tissue during caries removal. The findings provide a theoretical foundation and technical support for the future development of intelligent and standardized caries removal assistance systems, demonstrating significant clinical value and application potential. Materials and methods Dataset A total of 100 extracted human teeth with carious lesions were collected as research samples. During the in vitro mechanical caries removal process, a continuous photographic recording approach was employed. Specifically, for each tooth, 4 to 7 sequential images were captured throughout the caries excavation procedure to comprehensively document the progressive removal of carious tissue. This sequential imaging enabled the construction of a detailed visual representation of lesion changes during treatment, providing abundant and representative learning data for training the modified YOLOv8 model. Improved Algorithm Based on Multi-Scale Auxiliary Feature Fusion In this study, an improved small object detection algorithm based on multi-scale auxiliary feature fusion was proposed, built upon the YOLOv8 object detection framework. A dedicated multi-scale auxiliary feature extraction network was designed to capture shallow-level features, which were then fused with the backbone features. This integration ensures that even the deepest feature layers retain sufficient spatial information for accurate small object prediction. After iterative training, the optimized model was able to accurately detect and localize residual carious tissue. Dataset partitioning In this study, The collected extracted teeth were divided into training, validation, and test sets at a ratio of 7:1:2. The training set was used for training the deep learning model, the validation set for assessing model performance during training, and the test set for final evaluation after training completion. To effectively train and evaluate the modified YOLOv8 model—ensuring sufficient learning of data features during training and objective performance assessment during validation and testing—we enlisted three senior dental experts, each with over 10 years of clinical experience, to perform standardized annotation. Carious regions were then manually labeled on each image using the MakeSense platform ( https://www.makesense.ai ), and annotation files in multiple formats suitable for neural network training were generated. Model Training and Augmentation During the model training phase, to improve the model’s generalization ability and ensure robust detection performance across images captured from various angles and under different backgrounds, we collected caries images under both rubber dam and plaster model settings. A series of data augmentation techniques were applied to the training set, including but not limited to random cropping, rotation, scaling, flipping, and adjustments of color, contrast, and brightness. In addition, hard example mining was incorporated during training to enhance the model’s ability to detect small targets, such as residual carious tissue. Model Evaluation and Metrics After completing model training, we conducted a comprehensive evaluation using the test set to assess the model’s performance in detecting the presence of residual carious tissue in images. Key evaluation metrics included accuracy, sensitivity, and specificity. To further validate the model’s ability to accurately identify and localize carious sites, we also performed a statistical analysis of its performance in detecting specific lesion locations. Ethical considerations This study was conducted with the approval of the the Ethics Committee of Chongqing Dental Hospital (permission number: No. 2024YKYYLL005; permission date: November 11, 2024).All extracted teeth were collected with informed consent from donors or their legal representatives in accordance with institutional protocols. Results Basic Information of the Image Dataset A total of 100 extracted teeth with carious lesions were included in the study, and 666 images were collected in total, comprising 566 caries-positive images and 100 caries-free images. Among them, the training set contained 404 caries-positive and 70 caries-free images; the validation set included 50 caries-positive and 10 caries-free images; and the test set consisted of 112 caries-positive and 20 caries-free images. Detection Results Compared with the interpretations of the test set images by standardized and qualified dentists, the performance of the modified YOLOv8 model in detecting residual carious tissue is shown in Fig. 1 . The recognition results of the test set are shown in Table 1 . True positives (TP) refer to cases where both the model and the dentists identified the image as positive; false negatives (FN) refer to cases where the dentist identified the image as positive but the model identified it as negative; false positives (FP) refer to cases where the model identified the image as positive but the dentist identified it as negative; and true negatives (TN) refer to cases where both the model and the dentists identified the image as negative. Based on the data in Table 1 , various evaluation metrics of the model were calculated and are presented in Table 2 , including accuracy, sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), positive likelihood ratio (+ LR), and negative likelihood ratio (− LR). The figure shows the recognition of residual carious tissues under two backgrounds: rubber dam (A) and plaster cast B). For each background, three columns are displayed: (a) original image, (b) model prediction, and (c) expert annotation. Table 1 Detection Results of the Modified YOLOv8 Model for Residual Carious Tissue in the Test Set TP FN FP TN Residual Caries Status (Images) 99 13 3 17 Residual Caries Sites (n) 260 19 10 114 Table 2 Performance Evaluation Metrics of the Modified YOLOv8 Model Accuracy Sensitivity Specificity PPV NPV +LR - LR Residual Caries Status 87.9% 88.4% 85% 97.0% 56.7% 5.89 0.136 Residual Caries Sites 92.8% 93.2% 91.9% 96.3% 85.7% 11.506 0.068 Discussion Dental caries is one of the most prevalent oral diseases worldwide. Untreated carious lesions in permanent teeth remain a common global issue, significantly impacting oral health across populations [22]. The key to successful treatment lies in accurately assessing the extent and depth of carious tissue. Traditional caries detection methods primarily rely on dentists’ clinical experience and subjective judgment, which can limit the accuracy and consistency in determining the endpoint of caries removal. Variations in clinicians’ standards for complete caries excavation may result in residual pathogenic bacteria within the cavity, increasing the risk of secondary caries [23]. Achieving precise and thorough caries removal while preventing further progression and preserving pulp vitality remains a major challenge in clinical practice. Even for experienced clinicians, accurately removing all carious tissue without over-excavation is difficult, and complications such as accidental pulp exposure and recurrent caries are still frequently observed in practice. Compared with traditional manual diagnosis, AI has shown great potential in the field of dentistry. Due to its high precision, accuracy and sensitivity, AI has been gradually applied to oral radiology, orthodontics, prosthodontics, endodontics, implantology and other fields [24], which has greatly improved the level of diagnosis and treatment of dental diseases. AI can reduce the misdiagnosis and missed diagnosis caused by the lack of experience or subjective judgment of doctors. In recent years, the development of deep learning technology, especially Convolutional Neural Networks (CNN), has provided new possibilities for the automatic detection and diagnosis of dental diseases. YOLO is an object detection algorithm that represents a key breakthrough in the field of computer vision technology. This algorithm sets a new standard in terms of efficiency and accuracy, surpassing the performance of well-known methods such as R-CNN, Fast R-CNN, and Faster R-CNN [24–26]. YOLO series algorithm has the advantages of rapid detection, target recognition effect is better, detecting the advantages of good real-time performance, the proposed algorithm in various fields are widely used in target detection task, especially in the fields such as industry, agriculture, medicine and remote sensing [25]. In the study of James George and his colleagues, it was mentioned that YOLOv8 training model showed good results in the detection and classification of dental diseases such as dental caries, periodontal disease and oral cancer [26]. This study aims to investigate the detection capability of a modified YOLOv8 algorithm for residual carious tissue in extracted teeth, with the goal of enhancing the model’s ability for rapid identification and precise localization of residual caries. Likelihood ratios are commonly used evaluation metrics in medical diagnostic tests. The positive likelihood ratio (+ LR) represents the ratio of the true positive rate to the false positive rate; the larger the + LR, the higher the probability that a positive test result is a true positive. Conversely, the negative likelihood ratio (-LR) is the ratio of the false negative rate to the true negative rate; a smaller -LR indicates greater confidence in a negative test result. Generally, a test is considered effective when + LR exceeds 10 or -LR is below 0.1. According to the results of this study, the modified YOLOv8 model demonstrated favorable likelihood ratios for detecting carious sites, indicating that through learning from the training dataset, the model possesses strong discriminative power for homologous data. This further validates the effectiveness and potential of the improved YOLOv8 model in detecting residual carious tissue. In clinical caries removal procedures, a common observation is that carious lesions are often not confined to a single site. This is particularly evident in occlusal fissure caries, where even a single lesion may be segmented into multiple sites during the removal process. Therefore, to ultimately realize AI-assisted precision caries removal, it is essential that the model can accurately identify and consistently localize carious sites. This study not only evaluated the model’s ability to determine the presence or absence of residual carious tissue in images, but more importantly, assessed its capability to recognize and localize specific carious lesions. The results showed that the modified YOLOv8 model achieved an accuracy of 87.9%, sensitivity of 88.4%, and specificity of 85% in identifying the presence of residual caries in images. For carious site localization, the model achieved an accuracy of 92.8%, sensitivity of 93.2%, and specificity of 91.9%. These findings indicate that the system performs well in determining the existence of residual carious tissue and excels in the precise identification of carious sites. Moreover, we observed that in sequential images taken from the same tooth throughout the caries removal process, the model consistently recognized the same lesion with high stability. These promising results lay a solid foundation for improving future caries treatment. First, the system can enhance clinical efficiency by assisting clinicians in rapidly identifying residual caries through fast image analysis, thereby reducing treatment time and improving precision. Second, this research brings us closer to achieving AI-assisted, fully automated caries removal, which could further improve efficiency and reduce healthcare costs. This study also has certain limitations. First, it requires a high standard of image quality, as a well-annotated and high-quality dataset is one of the key factors affecting model performance. However, in real clinical practice, various unpredictable factors—such as water stains and blood—may interfere with image clarity and compromise recognition accuracy. Therefore, our next step is to further optimize the dataset and improve the model’s precision and robustness, enabling it to better adapt to real-world clinical scenarios.Second, most of the carious lesions included in this study were chronic caries, which typically exhibit more distinct discoloration, making them easier to detect. As a result, the model’s ability to recognize acute caries, which often present with less obvious visual cues, may require further enhancement.Finally, although the current model is capable of identifying the presence and location of residual carious tissue, its ability to precisely delineate the affected area still needs improvement. Further work is necessary to enhance the spatial accuracy of lesion localization. Conclusion The application of deep learning in caries detection presents new opportunities for the advancement of dental medicine. By leveraging object detection algorithms to automate the identification of residual carious tissue during the caries removal process, the reliance on clinicians’ experience can be significantly reduced. This approach minimizes the uncertainty caused by subjective judgments and enhances both the precision and efficiency of caries treatment.In this study, an improved YOLOv8 model was successfully applied to detect residual carious tissue in extracted teeth, achieving promising results. As the technology continues to evolve and its clinical applications expand, the enhanced YOLOv8 model is expected to play a more prominent role in dental practice, offering a highly accurate and efficient diagnostic support tool.The study confirms that the improved YOLOv8 model demonstrates good overall performance in recognizing residual carious tissue, laying a solid foundation for real-time detection and localization during the caries removal process. Ultimately, this work contributes to the development of AI-assisted automated caries removal in clinical dentistry. Declarations Ethics approval and consent to participate This study was reviewed and approved by the Ethics Committee of Chongqing Dental Hospital. Competing Interests The authors declare that they have no competing interests. Conflict of Interest Statement The authors declare that there are no known financial or non-financial conflicts of interest, or personal relationships, that could have appeared to influence the work reported in this manuscript. Funding This research was supported by the internal scientific research funding of Chongqing Dental Hospital. Author Contribution Ke Wang and Xulan Li were mainly responsible for image collection, model optimization and training, as well as manuscript drafting. Jing Lai and Fenfen Zhang contributed to the collection of extracted teeth and image acquisition. Yanming Chen was responsible for collecting the research results, performing statistical analysis, and drafting the initial manuscript. Senior dental experts Shanshan Guo, Xin Yu, and Gang Long were responsible for annotating the carious lesion sites. Xueman Wang supervised the overall project, including conceptual design, quality control, manuscript revision, and final approval. References GBD 2017 Oral Disorders Collaborators, Bernabe E, Marcenes W, et al. Global, Regional, A Systematic Analysis for the Global Burden of Disease 2017 Study.and National Levels and Trends in Burden of Oral Conditions from 1990 to 2017: J Dent Res. 2020; 99 (4) : 362–373. Askar H, Krois J, Gostemeyer G, et al. Secondary caries: .what is it, and how it can be controlled, detected, and managed? Clin Oral Investig. 2020; 724(5):1869–1876. Schwendicke F, Tzschoppe M, Paris S. Radiographic caries detection: 10.1016/j.j dent. 2021.103783]. A systematic review and meta-analysis [published correction appears in J Dent. 2021 Nov;114:103783. doi: J dent. 2015; 43 (8) : 924–933. Leal SC. Are standardised caries risk assessment models effective?. Evid Based Dent. 2018; 19 (4) : 102–103. Doi: 10.1038 / sj ebd. 6401338 Correa-Faria P, Paixao-Goncalves S, Ramos-Jorge ML, Paiva SM, Case-control study.Pordeus IA. Developmental enamel defects are associated with early childhood caries: Int J Paediatr Dent. 2020; 30(1):11–17. Kim J, Lee HS, Song IS, Jung KH. DeNTNet: Deep Neural Transfer Network for the detection of periodontal bone loss using panoramic dental radiographs. Sci Rep. 2019; 9(1):17615. Published 2019 Nov 26. Chen X, Wang X, Zhang K, et al. Recent advances and clinical applications of deep learning in medical image analysis. Med Image Anal. 2022; 79:102444. [8].Matsuo Y, LeCun Y, Sahan. M, et al. Deep learning, reinforcement learning, and world models. Neural Netw. 2022; 152:267–275. Kuhnisch J, Meyer O, Hesenius M, Hickel R, Gruhn V. Caries Detection on Intraoral Images Using Artificial Intelligence. J Dent Res101 (2) : 158–165. Askar H, Krois J, Rohrer C, et al. Detecting white spot lesions on dental photography using deep learning: A pilot study. J Dent. 2021; 107:103615. 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 h. 2022; Healt22(1):573. Published 2022 Dec 7. Lee JH, Kim DH, Jeong SN, Choi SH. Detection and diagnosis of dental caries using a deep learning-based convolutional neural network algorithm. J t. 2018; Den77:106–111. Hung M, Voss MW, Rosales MN, et al. Application of machine learning for diagnostic prediction of root caries. Gerodontology. 2019; 36(4):395–404. [14].Cantu AG, Gehrung S, Krois J, et al. Detecting caries lesions of different radiographic extension on bitewings using deep learning. J Dent. 2020; 100:103425. Patil S, Albogami S, Hosmani J, et al. Artificial Intelligence in the Diagnosis of Oral Diseases: Applications and Pitfalls.). 2022; Diagnostics (Basel12 (5) : 1029. Published 2022 Apr 19. Asiri AF, Altuwalah AS. The role of neural artificial intelligence for diagnosis and treatment planning in endodontics: A qualitative review. Saudi Dent J. 2022; 34 (4) : 270–281. Nordblom NF, Buttner M, Schwendicke F. Artificial Intelligence in Orthodontics: Critical Review. J Dent Res.2024; 103 (6) : 577–584. Rawal S. Guided innovations: Robot-assisted dental implant surgery. J Prosthet Dent. 2022; 127 (5) : 673–674. Setzer FC, Li J, Khan AA. The Use of Artificial Intelligence in Endodontics. J Dent Res. 2024; 103 (9) : 853–862. [20].Parveen Rahamathulla M, Sam Emmanuel WR, Bindhu A, Mustaq Ahmed M. YOLOv8's advancements in tuberculosis identification from chest images. Front Big Data. 2024;7:1401981. Published 2024 Jun 27. Kaur A, Jyoti D, Sharma A, Yelam D, Goyal R, Nath A. Deep caries detection using deep learning: from dataset acquisition to detection. Clin Oral Investig. 2024;28(12):677. Published 2024 Dec 2. Kassebaum NJ, Bernabe E, Dahiya M, Bhandari B, Murray CJ, Marcenes W. Global burden of untreated caries: a systematic review and metaregression. J Dent Res. 2015; 94 (5) : 650–658. Askar H, Krois J, Gostemeyer G, et al. Secondary caries: .what is it, and how it can be controlled, detected, and managed? Clin Oral Investig. 2020; 24(5):1869–1876. Schwendicke F, Samek W, Krois J. Artificial Intelligence in Dentistry: Chances and Challenges. J Dent Res. 2020; 99 (7) : 769–774. Diwan T, Anirudh G, Tembhurne JV. Object detection using YOLO: challenges, architectural successors, datasets and applications. Multimed Tools Appl. 2023; 82(6):9243–9275. George, James, T. S. Hemanth, Joshua Raju, Johan George Mattapallil, and N. Naveen. "Dental radiography analysis and diagnosis using YOLOv8." In 2023 9th International Conference on Smart Computing and Communications (ICSCC), pp. 102–107. IEEE, 2023. Additional Declarations No competing interests reported. Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-6933790","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":507893331,"identity":"a218893e-4063-4546-91df-c757fc8354f8","order_by":0,"name":"Ke Wang","email":"","orcid":"","institution":"Chongqing Dental Hospital","correspondingAuthor":false,"prefix":"","firstName":"Ke","middleName":"","lastName":"Wang","suffix":""},{"id":507893332,"identity":"11e28df1-a212-4030-87e7-47027469ddae","order_by":1,"name":"Xulan Li","email":"","orcid":"","institution":"Chongqing Dental Hospital","correspondingAuthor":false,"prefix":"","firstName":"Xulan","middleName":"","lastName":"Li","suffix":""},{"id":507893333,"identity":"15754518-8bff-4fad-8aab-6ce9a885b1e9","order_by":2,"name":"Jing Lai","email":"","orcid":"","institution":"Chongqing Dental Hospital","correspondingAuthor":false,"prefix":"","firstName":"Jing","middleName":"","lastName":"Lai","suffix":""},{"id":507893334,"identity":"f2865ad2-9d29-48a5-98c2-2debce324565","order_by":3,"name":"Shanshan Guo","email":"","orcid":"","institution":"Chongqing Dental Hospital","correspondingAuthor":false,"prefix":"","firstName":"Shanshan","middleName":"","lastName":"Guo","suffix":""},{"id":507893335,"identity":"799ad568-bec1-44b7-825a-d39951332b88","order_by":4,"name":"Yanming Chen","email":"","orcid":"","institution":"Chongqing Dental Hospital","correspondingAuthor":false,"prefix":"","firstName":"Yanming","middleName":"","lastName":"Chen","suffix":""},{"id":507893336,"identity":"6af4f36c-e1d4-4231-8416-b14fe20c69d1","order_by":5,"name":"Fenfen Zhang","email":"","orcid":"","institution":"Chongqing Dental Hospital","correspondingAuthor":false,"prefix":"","firstName":"Fenfen","middleName":"","lastName":"Zhang","suffix":""},{"id":507893337,"identity":"8526255c-fd75-4975-a52c-56714350ea28","order_by":6,"name":"Xin Yu","email":"","orcid":"","institution":"Chongqing Dental Hospital","correspondingAuthor":false,"prefix":"","firstName":"Xin","middleName":"","lastName":"Yu","suffix":""},{"id":507893338,"identity":"f8b0635f-fa17-49b8-a7a7-06ed543d07a7","order_by":7,"name":"Gang Long","email":"","orcid":"","institution":"Chongqing Dental Hospital","correspondingAuthor":false,"prefix":"","firstName":"Gang","middleName":"","lastName":"Long","suffix":""},{"id":507893339,"identity":"032763b9-e94d-43cd-9ef3-cb39e604ee3c","order_by":8,"name":"Xueman Wang","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAxUlEQVRIiWNgGAWjYBAC+/sPG4z/VNTYMbY3EKvnQHJDAc+ZY8nMPQeI1pLe8IG3jZmxfUYCkToYGw42bpA4w8bMO/PxxhsMNTbRBLUwMzY2GxhUyPBJzk4rtmA4lpbbQEgLGzNjm0EC0BbD2TlmEowNhwlr4WFjbP9xEOiX/TfPEKlFgoexwbARqKVxBg+RWgyAyowZgIHM2AP0SwIxfjGQYH9gzACOysMbb3yosSGsBVV7AinKIVpI1TEKRsEoGAUjAwAAE8FATPR2bfMAAAAASUVORK5CYII=","orcid":"","institution":"Chongqing Dental Hospital","correspondingAuthor":true,"prefix":"","firstName":"Xueman","middleName":"","lastName":"Wang","suffix":""}],"badges":[],"createdAt":"2025-06-19 21:53:13","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-6933790/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-6933790/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":90470846,"identity":"5debb2d2-19a6-450d-97dc-75bd5df7b8e6","added_by":"auto","created_at":"2025-09-03 06:16:30","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":1019663,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eDetection results of the improved YOLOv8 model under different backgrounds.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe figure shows the recognition of residual carious tissues under two backgrounds: rubber dam (A) and plaster cast B). For each background, three columns are displayed: (a) original image, (b) model prediction, and (c)expert annotation.\u003c/p\u003e","description":"","filename":"floatimage16.png","url":"https://assets-eu.researchsquare.com/files/rs-6933790/v1/b92efed71f579e639610ba86.png"},{"id":99797182,"identity":"6da81dbf-41eb-4d70-a51c-782083ca814b","added_by":"auto","created_at":"2026-01-08 13:45:23","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1941189,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6933790/v1/74707280-3357-468f-8fec-a2891307bfc2.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Intelligent Detection of Residual Carious Tissue During Caries Removal Using an Improved YOLOv8 Model","fulltext":[{"header":"Introduction","content":"\u003cp\u003eDental caries is one of the most prevalent oral diseases worldwide [1], significantly impacting individuals\u0026rsquo; oral health and overall quality of life. In the treatment of dental caries, the complete removal of decayed tissue is directly related to both therapeutic efficacy and long-term prognosis [2]. Currently, most clinicians rely primarily on direct visual inspection and tactile examination using a dental probe to assess the color and texture of residual tissue [3]. This approach is inherently subjective and heavily dependent on the operator\u0026rsquo;s clinical experience, lacking standardization and objectivity.\u003c/p\u003e\u003cp\u003eWith the continuous advancement of artificial intelligence (AI), the digitalization and automation of diagnostic and therapeutic procedures have become a major trend in modern medicine [4,5]. At present, AI technologies\u0026mdash;particularly deep learning\u0026mdash;can simulate the human brain\u0026rsquo;s ability to process medical images through multilayer neural networks [6\u0026ndash;8]. Meanwhile, computer vision enables machines to understand and interpret visual information in images and videos [9,10]. For example, image segmentation techniques can isolate teeth and gingiva from the background to facilitate accurate health assessment [11], while object detection algorithms can identify specific targets within images. These automated processes not only improve the efficiency of image analysis but also reduce human error, enhancing the reliability of clinical decision-making.\u003c/p\u003e\u003cp\u003eIn recent years, AI has made rapid progress in the field of dentistry, demonstrating promising applications in diagnostic assistance, treatment planning, and beyond [12\u0026ndash;15]. AI-assisted precision procedures have also begun to find practical use in clinical settings [16\u0026ndash;19]. As a result, the development of an AI system capable of real-time guidance and localization of residual carious tissue during caries removal holds great potential. Such a system could improve the quality of operative procedures and lay the groundwork for the future realization of automated caries removal, marking a significant advancement in the application of AI in operative dentistry and endodontics.\u003c/p\u003e\u003cp\u003eTo achieve this goal, the primary technical challenge is enabling rapid and accurate identification and localization of residual carious tissue during dynamic operative procedures. This would allow the task of intraoperative assessment, traditionally carried out by clinicians, to be assisted or even replaced by an AI system.\u003c/p\u003e\u003cp\u003eYOLOv8(You Only Look Once, Version 8), as a new-generation object detection algorithm, offers faster computational speed and higher detection accuracy[20,21]. Its application in identifying residual carious tissue during tooth preparation holds promise for enhancing the intelligence level of dental diagnosis and treatment, enabling real-time assistance in localization and assessment. This study explores the feasibility of using an improved YOLOv8 model to detect and locate residual carious tissue during caries removal. The findings provide a theoretical foundation and technical support for the future development of intelligent and standardized caries removal assistance systems, demonstrating significant clinical value and application potential.\u003c/p\u003e"},{"header":"Materials and methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e\u003ch2\u003eDataset\u003c/h2\u003e\u003cp\u003eA total of 100 extracted human teeth with carious lesions were collected as research samples. During the in vitro mechanical caries removal process, a continuous photographic recording approach was employed. Specifically, for each tooth, 4 to 7 sequential images were captured throughout the caries excavation procedure to comprehensively document the progressive removal of carious tissue. This sequential imaging enabled the construction of a detailed visual representation of lesion changes during treatment, providing abundant and representative learning data for training the modified YOLOv8 model.\u003c/p\u003e\u003c/div\u003e\n\u003ch3\u003eImproved Algorithm Based on Multi-Scale Auxiliary Feature Fusion\u003c/h3\u003e\n\u003cp\u003eIn this study, an improved small object detection algorithm based on multi-scale auxiliary feature fusion was proposed, built upon the YOLOv8 object detection framework. A dedicated multi-scale auxiliary feature extraction network was designed to capture shallow-level features, which were then fused with the backbone features. This integration ensures that even the deepest feature layers retain sufficient spatial information for accurate small object prediction. After iterative training, the optimized model was able to accurately detect and localize residual carious tissue.\u003c/p\u003e\n\u003ch3\u003eDataset partitioning\u003c/h3\u003e\n\u003cp\u003eIn this study, The collected extracted teeth were divided into training, validation, and test sets at a ratio of 7:1:2. The training set was used for training the deep learning model, the validation set for assessing model performance during training, and the test set for final evaluation after training completion. To effectively train and evaluate the modified YOLOv8 model\u0026mdash;ensuring sufficient learning of data features during training and objective performance assessment during validation and testing\u0026mdash;we enlisted three senior dental experts, each with over 10 years of clinical experience, to perform standardized annotation. Carious regions were then manually labeled on each image using the MakeSense platform (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.makesense.ai\u003c/span\u003e\u003cspan address=\"https://www.makesense.ai\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e), and annotation files in multiple formats suitable for neural network training were generated.\u003c/p\u003e\n\u003ch3\u003eModel Training and Augmentation\u003c/h3\u003e\n\u003cp\u003eDuring the model training phase, to improve the model\u0026rsquo;s generalization ability and ensure robust detection performance across images captured from various angles and under different backgrounds, we collected caries images under both rubber dam and plaster model settings. A series of data augmentation techniques were applied to the training set, including but not limited to random cropping, rotation, scaling, flipping, and adjustments of color, contrast, and brightness. In addition, hard example mining was incorporated during training to enhance the model\u0026rsquo;s ability to detect small targets, such as residual carious tissue.\u003c/p\u003e\n\u003ch3\u003eModel Evaluation and Metrics\u003c/h3\u003e\n\u003cp\u003eAfter completing model training, we conducted a comprehensive evaluation using the test set to assess the model\u0026rsquo;s performance in detecting the presence of residual carious tissue in images. Key evaluation metrics included accuracy, sensitivity, and specificity. To further validate the model\u0026rsquo;s ability to accurately identify and localize carious sites, we also performed a statistical analysis of its performance in detecting specific lesion locations.\u003c/p\u003e\u003cdiv id=\"Sec8\" class=\"Section2\"\u003e\u003ch2\u003eEthical considerations\u003c/h2\u003e\u003cp\u003eThis study was conducted with the approval of the the Ethics Committee of Chongqing Dental Hospital (permission number: No. 2024YKYYLL005; permission date: November 11, 2024).All extracted teeth were collected with informed consent from donors or their legal representatives in accordance with institutional protocols.\u003c/p\u003e\u003c/div\u003e"},{"header":"Results","content":"\u003cdiv id=\"Sec10\" class=\"Section2\"\u003e\u003ch2\u003eBasic Information of the Image Dataset\u003c/h2\u003e\u003cp\u003eA total of 100 extracted teeth with carious lesions were included in the study, and 666 images were collected in total, comprising 566 caries-positive images and 100 caries-free images. Among them, the training set contained 404 caries-positive and 70 caries-free images; the validation set included 50 caries-positive and 10 caries-free images; and the test set consisted of 112 caries-positive and 20 caries-free images.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec11\" class=\"Section2\"\u003e\u003ch2\u003eDetection Results\u003c/h2\u003e\u003cp\u003eCompared with the interpretations of the test set images by standardized and qualified dentists, the performance of the modified YOLOv8 model in detecting residual carious tissue is shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e.\u003c/p\u003e\u003cp\u003eThe recognition results of the test set are shown in Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e. True positives (TP) refer to cases where both the model and the dentists identified the image as positive; false negatives (FN) refer to cases where the dentist identified the image as positive but the model identified it as negative; false positives (FP) refer to cases where the model identified the image as positive but the dentist identified it as negative; and true negatives (TN) refer to cases where both the model and the dentists identified the image as negative. Based on the data in Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e, various evaluation metrics of the model were calculated and are presented in Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e, including accuracy, sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), positive likelihood ratio (+\u0026thinsp;LR), and negative likelihood ratio (\u0026minus;\u0026thinsp;LR).\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003eThe figure shows the recognition of residual carious tissues under two backgrounds: rubber dam (A) and plaster cast B). For each background, three columns are displayed: (a) original image, (b) model prediction, and (c) expert annotation.\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eDetection Results of the Modified YOLOv8 Model for Residual Carious Tissue in the Test Set\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"5\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eTP\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eFN\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eFP\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\u003e\u003cp\u003eTN\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eResidual Caries Status (Images)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e99\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e13\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e3\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e17\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eResidual Caries Sites (n)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e260\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e19\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e10\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e114\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003ePerformance Evaluation Metrics of the Modified YOLOv8 Model\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"8\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c8\" colnum=\"8\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eAccuracy\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eSensitivity\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eSpecificity\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\u003e\u003cp\u003ePPV\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c6\"\u003e\u003cp\u003eNPV\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c7\"\u003e\u003cp\u003e+LR\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c8\"\u003e\u003cp\u003e- LR\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eResidual Caries Status\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e87.9%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e88.4%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e85%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e97.0%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e56.7%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e5.89\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e\u003cp\u003e0.136\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eResidual Caries Sites\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e92.8%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e93.2%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e91.9%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e96.3%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e85.7%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e11.506\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e\u003cp\u003e0.068\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003c/div\u003e"},{"header":"Discussion","content":"\u003cp\u003eDental caries is one of the most prevalent oral diseases worldwide. Untreated carious lesions in permanent teeth remain a common global issue, significantly impacting oral health across populations [22]. The key to successful treatment lies in accurately assessing the extent and depth of carious tissue. Traditional caries detection methods primarily rely on dentists\u0026rsquo; clinical experience and subjective judgment, which can limit the accuracy and consistency in determining the endpoint of caries removal. Variations in clinicians\u0026rsquo; standards for complete caries excavation may result in residual pathogenic bacteria within the cavity, increasing the risk of secondary caries [23]. Achieving precise and thorough caries removal while preventing further progression and preserving pulp vitality remains a major challenge in clinical practice. Even for experienced clinicians, accurately removing all carious tissue without over-excavation is difficult, and complications such as accidental pulp exposure and recurrent caries are still frequently observed in practice.\u003c/p\u003e\u003cp\u003eCompared with traditional manual diagnosis, AI has shown great potential in the field of dentistry. Due to its high precision, accuracy and sensitivity, AI has been gradually applied to oral radiology, orthodontics, prosthodontics, endodontics, implantology and other fields [24], which has greatly improved the level of diagnosis and treatment of dental diseases. AI can reduce the misdiagnosis and missed diagnosis caused by the lack of experience or subjective judgment of doctors. In recent years, the development of deep learning technology, especially Convolutional Neural Networks (CNN), has provided new possibilities for the automatic detection and diagnosis of dental diseases. YOLO is an object detection algorithm that represents a key breakthrough in the field of computer vision technology. This algorithm sets a new standard in terms of efficiency and accuracy, surpassing the performance of well-known methods such as R-CNN, Fast R-CNN, and Faster R-CNN [24\u0026ndash;26]. YOLO series algorithm has the advantages of rapid detection, target recognition effect is better, detecting the advantages of good real-time performance, the proposed algorithm in various fields are widely used in target detection task, especially in the fields such as industry, agriculture, medicine and remote sensing [25]. In the study of James George and his colleagues, it was mentioned that YOLOv8 training model showed good results in the detection and classification of dental diseases such as dental caries, periodontal disease and oral cancer [26].\u003c/p\u003e\u003cp\u003eThis study aims to investigate the detection capability of a modified YOLOv8 algorithm for residual carious tissue in extracted teeth, with the goal of enhancing the model\u0026rsquo;s ability for rapid identification and precise localization of residual caries. Likelihood ratios are commonly used evaluation metrics in medical diagnostic tests. The positive likelihood ratio (+\u0026thinsp;LR) represents the ratio of the true positive rate to the false positive rate; the larger the +\u0026thinsp;LR, the higher the probability that a positive test result is a true positive. Conversely, the negative likelihood ratio (-LR) is the ratio of the false negative rate to the true negative rate; a smaller -LR indicates greater confidence in a negative test result. Generally, a test is considered effective when +\u0026thinsp;LR exceeds 10 or -LR is below 0.1. According to the results of this study, the modified YOLOv8 model demonstrated favorable likelihood ratios for detecting carious sites, indicating that through learning from the training dataset, the model possesses strong discriminative power for homologous data. This further validates the effectiveness and potential of the improved YOLOv8 model in detecting residual carious tissue.\u003c/p\u003e\u003cp\u003eIn clinical caries removal procedures, a common observation is that carious lesions are often not confined to a single site. This is particularly evident in occlusal fissure caries, where even a single lesion may be segmented into multiple sites during the removal process. Therefore, to ultimately realize AI-assisted precision caries removal, it is essential that the model can accurately identify and consistently localize carious sites.\u003c/p\u003e\u003cp\u003eThis study not only evaluated the model\u0026rsquo;s ability to determine the presence or absence of residual carious tissue in images, but more importantly, assessed its capability to recognize and localize specific carious lesions.\u003c/p\u003e\u003cp\u003eThe results showed that the modified YOLOv8 model achieved an accuracy of 87.9%, sensitivity of 88.4%, and specificity of 85% in identifying the presence of residual caries in images. For carious site localization, the model achieved an accuracy of 92.8%, sensitivity of 93.2%, and specificity of 91.9%. These findings indicate that the system performs well in determining the existence of residual carious tissue and excels in the precise identification of carious sites.\u003c/p\u003e\u003cp\u003eMoreover, we observed that in sequential images taken from the same tooth throughout the caries removal process, the model consistently recognized the same lesion with high stability. These promising results lay a solid foundation for improving future caries treatment. First, the system can enhance clinical efficiency by assisting clinicians in rapidly identifying residual caries through fast image analysis, thereby reducing treatment time and improving precision. Second, this research brings us closer to achieving AI-assisted, fully automated caries removal, which could further improve efficiency and reduce healthcare costs.\u003c/p\u003e\u003cp\u003eThis study also has certain limitations. First, it requires a high standard of image quality, as a well-annotated and high-quality dataset is one of the key factors affecting model performance. However, in real clinical practice, various unpredictable factors\u0026mdash;such as water stains and blood\u0026mdash;may interfere with image clarity and compromise recognition accuracy. Therefore, our next step is to further optimize the dataset and improve the model\u0026rsquo;s precision and robustness, enabling it to better adapt to real-world clinical scenarios.Second, most of the carious lesions included in this study were chronic caries, which typically exhibit more distinct discoloration, making them easier to detect. As a result, the model\u0026rsquo;s ability to recognize acute caries, which often present with less obvious visual cues, may require further enhancement.Finally, although the current model is capable of identifying the presence and location of residual carious tissue, its ability to precisely delineate the affected area still needs improvement. Further work is necessary to enhance the spatial accuracy of lesion localization.\u003c/p\u003e"},{"header":"Conclusion","content":"\u003cp\u003eThe application of deep learning in caries detection presents new opportunities for the advancement of dental medicine. By leveraging object detection algorithms to automate the identification of residual carious tissue during the caries removal process, the reliance on clinicians\u0026rsquo; experience can be significantly reduced. This approach minimizes the uncertainty caused by subjective judgments and enhances both the precision and efficiency of caries treatment.In this study, an improved YOLOv8 model was successfully applied to detect residual carious tissue in extracted teeth, achieving promising results. As the technology continues to evolve and its clinical applications expand, the enhanced YOLOv8 model is expected to play a more prominent role in dental practice, offering a highly accurate and efficient diagnostic support tool.The study confirms that the improved YOLOv8 model demonstrates good overall performance in recognizing residual carious tissue, laying a solid foundation for real-time detection and localization during the caries removal process. Ultimately, this work contributes to the development of AI-assisted automated caries removal in clinical dentistry.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eEthics approval and consent to participate\u003c/strong\u003e\u003cp\u003e This study was reviewed and approved by the Ethics Committee of Chongqing Dental Hospital.\u003c/p\u003e\u003c/p\u003e\u003cp\u003e\u003ch2\u003eCompeting Interests\u003c/h2\u003e\u003cp\u003eThe authors declare that they have no competing interests.\u003c/p\u003e\u003c/p\u003e\u003cp\u003e\u003cstrong\u003eConflict of Interest Statement\u003c/strong\u003e\u003cp\u003eThe authors declare that there are no known financial or non-financial conflicts of interest, or personal relationships, that could have appeared to influence the work reported in this manuscript.\u003c/p\u003e\u003c/p\u003e\u003ch2\u003eFunding\u003c/h2\u003e\u003cp\u003eThis research was supported by the internal scientific research funding of Chongqing Dental Hospital.\u003c/p\u003e\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eKe Wang and Xulan Li were mainly responsible for image collection, model optimization and training, as well as manuscript drafting. Jing Lai and Fenfen Zhang contributed to the collection of extracted teeth and image acquisition. Yanming Chen was responsible for collecting the research results, performing statistical analysis, and drafting the initial manuscript. Senior dental experts Shanshan Guo, Xin Yu, and Gang Long were responsible for annotating the carious lesion sites. Xueman Wang supervised the overall project, including conceptual design, quality control, manuscript revision, and final approval.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eGBD 2017 Oral Disorders Collaborators, Bernabe E, Marcenes W, et al. Global, Regional, A Systematic Analysis for the Global Burden of Disease 2017 Study.and National Levels and Trends in Burden of Oral Conditions from 1990 to 2017: J Dent Res. 2020; 99 (4) : 362\u0026ndash;373.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eAskar H, Krois J, Gostemeyer G, et al. Secondary caries: .what is it, and how it can be controlled, detected, and managed? Clin Oral Investig. 2020; 724(5):1869\u0026ndash;1876.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSchwendicke F, Tzschoppe M, Paris S. Radiographic caries detection: 10.1016/j.j dent. 2021.103783]. A systematic review and meta-analysis [published correction appears in J Dent. 2021 Nov;114:103783. doi: J dent. 2015; 43 (8) : 924\u0026ndash;933.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLeal SC. Are standardised caries risk assessment models effective?. Evid Based Dent. 2018; 19 (4) : 102\u0026ndash;103. Doi: 10.1038 / sj ebd. 6401338\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eCorrea-Faria P, Paixao-Goncalves S, Ramos-Jorge ML, Paiva SM, Case-control study.Pordeus IA. Developmental enamel defects are associated with early childhood caries: Int J Paediatr Dent. 2020; 30(1):11\u0026ndash;17.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKim J, Lee HS, Song IS, Jung KH. DeNTNet: Deep Neural Transfer Network for the detection of periodontal bone loss using panoramic dental radiographs. Sci Rep. 2019; 9(1):17615. Published 2019 Nov 26.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eChen X, Wang X, Zhang K, et al. Recent advances and clinical applications of deep learning in medical image analysis. Med Image Anal. 2022; 79:102444. [8].Matsuo Y, LeCun Y, Sahan. M, et al. Deep learning, reinforcement learning, and world models. Neural Netw. 2022; 152:267\u0026ndash;275.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKuhnisch J, Meyer O, Hesenius M, Hickel R, Gruhn V. Caries Detection on Intraoral Images Using Artificial Intelligence. J Dent Res101 (2) : 158\u0026ndash;165.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eAskar H, Krois J, Rohrer C, et al. Detecting white spot lesions on dental photography using deep learning: A pilot study. J Dent. 2021; 107:103615.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003ePark EY, Cho H, Kang S, Jeong S, Kim EK. Caries detection with tooth surface segmentation on intraoral photographic images using deep learning. BMC Oral h. 2022; Healt22(1):573. Published 2022 Dec 7.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLee JH, Kim DH, Jeong SN, Choi SH. Detection and diagnosis of dental caries using a deep learning-based convolutional neural network algorithm. J t. 2018; Den77:106\u0026ndash;111.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHung M, Voss MW, Rosales MN, et al. Application of machine learning for diagnostic prediction of root caries. Gerodontology. 2019; 36(4):395\u0026ndash;404. [14].Cantu AG, Gehrung S, Krois J, et al. Detecting caries lesions of different radiographic extension on bitewings using deep learning. J Dent. 2020; 100:103425.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003ePatil S, Albogami S, Hosmani J, et al. Artificial Intelligence in the Diagnosis of Oral Diseases: Applications and Pitfalls.). 2022; Diagnostics (Basel12 (5) : 1029. Published 2022 Apr 19.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eAsiri AF, Altuwalah AS. The role of neural artificial intelligence for diagnosis and treatment planning in endodontics: A qualitative review. Saudi Dent J. 2022; 34 (4) : 270\u0026ndash;281.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eNordblom NF, Buttner M, Schwendicke F. Artificial Intelligence in Orthodontics: Critical Review. J Dent Res.2024; 103 (6) : 577\u0026ndash;584.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eRawal S. Guided innovations: Robot-assisted dental implant surgery. J Prosthet Dent. 2022; 127 (5) : 673\u0026ndash;674.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSetzer FC, Li J, Khan AA. The Use of Artificial Intelligence in Endodontics. J Dent Res. 2024; 103 (9) : 853\u0026ndash;862. [20].Parveen Rahamathulla M, Sam Emmanuel WR, Bindhu A, Mustaq Ahmed M. YOLOv8's advancements in tuberculosis identification from chest images. Front Big Data. 2024;7:1401981. Published 2024 Jun 27.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKaur A, Jyoti D, Sharma A, Yelam D, Goyal R, Nath A. Deep caries detection using deep learning: from dataset acquisition to detection. Clin Oral Investig. 2024;28(12):677. Published 2024 Dec 2.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKassebaum NJ, Bernabe E, Dahiya M, Bhandari B, Murray CJ, Marcenes W. Global burden of untreated caries: a systematic review and metaregression. J Dent Res. 2015; 94 (5) : 650\u0026ndash;658.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eAskar H, Krois J, Gostemeyer G, et al. Secondary caries: .what is it, and how it can be controlled, detected, and managed? Clin Oral Investig. 2020; 24(5):1869\u0026ndash;1876.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSchwendicke F, Samek W, Krois J. Artificial Intelligence in Dentistry: Chances and Challenges. J Dent Res. 2020; 99 (7) : 769\u0026ndash;774.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eDiwan T, Anirudh G, Tembhurne JV. Object detection using YOLO: challenges, architectural successors, datasets and applications. Multimed Tools Appl. 2023; 82(6):9243\u0026ndash;9275.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eGeorge, James, T. S. Hemanth, Joshua Raju, Johan George Mattapallil, and N. Naveen. \"Dental radiography analysis and diagnosis using YOLOv8.\" In 2023 9th International Conference on Smart Computing and Communications (ICSCC), pp. 102\u0026ndash;107. IEEE, 2023.\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"YOLOv8, Deep learning, Dental caries, Residual carious tissue","lastPublishedDoi":"10.21203/rs.3.rs-6933790/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-6933790/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003e\u003cb\u003eObjective\u003c/b\u003e:\u003c/p\u003e\u003cp\u003eThis study aimed to develop a prototype of an AI-based recognition system for identifying residual carious tissue during caries removal by improving the YOLOv8 model. The goal was to provide preliminary theoretical and technical support for real-time AI-assisted guidance and localization of residual caries.\u003c/p\u003e\u003cp\u003e\u003cb\u003eMethods\u003c/b\u003e:\u003c/p\u003e\u003cp\u003eA total of 100 extracted teeth with carious lesions were collected, numbered, and randomly assigned to training, validation, and test sets at a ratio of 7:1:2. During the extracoronal caries removal process for each tooth, 4 to 7 consecutive images were captured until all carious tissue was removed. Carious regions in the training images were manually annotated and used to train the improved YOLOv8 model. The model's predictions on the test set were compared with those made by standardized, experienced dental professionals. Performance was evaluated using accuracy, sensitivity, and specificity.\u003c/p\u003e\u003cp\u003e\u003cb\u003eResults\u003c/b\u003e:\u003c/p\u003e\u003cp\u003eA total of 666 images were collected, including 566 images with carious lesions and 100 without. Compared with expert annotations, the improved YOLOv8 model achieved an accuracy of 87.9%, sensitivity of 88.4%, and specificity of 85% in identifying the presence of residual caries. For the localization of specific residual carious sites, the model achieved an accuracy of 92.8%, sensitivity of 93.2%, and specificity of 91.9%.\u003c/p\u003e\u003cp\u003e\u003cb\u003eConclusion\u003c/b\u003e:\u003c/p\u003e\u003cp\u003eThe improved YOLOv8 model demonstrated strong capability in detecting and localizing residual carious tissue across consecutive images with high stability, indicating promising potential for clinical application.\u003c/p\u003e","manuscriptTitle":"Intelligent Detection of Residual Carious Tissue During Caries Removal Using an Improved YOLOv8 Model","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-09-03 06:16:25","doi":"10.21203/rs.3.rs-6933790/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"cccef21a-a4f7-4e5b-807c-3d5003be4258","owner":[],"postedDate":"September 3rd, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2026-01-07T17:39:01+00:00","versionOfRecord":[],"versionCreatedAt":"2025-09-03 06:16:25","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-6933790","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-6933790","identity":"rs-6933790","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}