Artificial intelligence-enhanced diagnosis of temporomandibular joint osteoarthritis using temporomandibular joint panoramic radiography and joint noise data

Artificial intelligence-enhanced diagnosis of temporomandibular joint osteoarthritis using temporomandibular joint panoramic radiography and joint noise data | 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 Article Artificial intelligence-enhanced diagnosis of temporomandibular joint osteoarthritis using temporomandibular joint panoramic radiography and joint noise data Eunhye Choi, Seokwon Shin, Kijin Lee, Taejin An, Richard K. Lee, and 3 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-5086242/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 13 Jan, 2025 Read the published version in Scientific Reports → Version 1 posted 11 You are reading this latest preprint version Abstract This study aimed to develop an artificial intelligence (AI) model for the screening of temporomandibular joint osteoarthritis (TMJ OA) using temporomandibular joint (TMJ) panoramic radiography and joint noise data. A total of 2,631 TMJ panoramic images were collected, resulting in a final dataset of 3,908 images (2,127 normal (N) and 1,781 TMJ OA (D)) after excluding indeterminate cases and errors. AI models using GoogleNet were evaluated with six different combinations of image data, clinician-detected crepitus, and patient-reported joint noise. The model that integrated all joint noise data with imaging, demonstrated the highest performance, achieving an F1-score of 0.72. Another model, which incorporated both imaging and crepitus, also achieved the same F1-score but had lower D recall (0.55 vs. 0.67) and N precision (0.71 vs. 0.74). The AI models outperformed orofacial pain specialists when provided with imaging alone or in combination with all joint noise data. These findings suggest that AI-enhanced TMJ OA diagnosis using TMJ panoramic radiography and joint noise data offers a promising approach for early detection and improved patient care. The results underscore AI's capability to integrate diverse diagnostic factors, providing a comprehensive and accurate assessment that surpasses traditional methods. Health sciences/Diseases Health sciences/Medical research Temporomandibular joint osteoarthritis Artificial intelligence Temporomandibular joint panoramic radiography Joint noise Figures Figure 1 Figure 2 Figure 3 Figure 4 Introduction Temporomandibular joint osteoarthritis (TMJ OA) is a degenerative condition characterized by cartilage breakdown and bone deformation in the mandibular condyle and articular fossa, significantly affecting patients' quality of life by impairing function and aesthetics 1 . Interestingly, while osteoarthritis (OA) is typically considered female predominant and age-related, the linear correlation between bone change and age observed in other joints is not evident in the temporomandibular joint (TMJ). Temporomandibular disorder (TMD) is most prevalent in young adults aged 20 to 40, and TMJ OA may begin at an early age 2 . Notably, the severity of TMJ OA does not always correspond with pain intensity; TMJ OA changes in young Korean TMD patients are as common as in older individuals, showing no correlation with clinical pain or disc displacement 3 . This makes early detection both challenging and essential, especially since significant degeneration often occurs without overt symptoms. Cone-beam computed tomography (CBCT) is the reference standard for diagnosing TMJ OA due to its high diagnostic accuracy and detailed anatomical assessment 4 . However, CBCT is costly and exposes patients to higher radiation doses compared to other imaging modalities. While panoramic radiography is more accessible and cost-effective, it lacks comprehensive evaluation capabilities for the TMJ region. Conversely, TMJ panoramic radiography provides a more targeted evaluation, enabling assessment of the mandibular condyle's range of motion without significant projection limitations. Recent a systematic review and meta-analysis of artificial intelligence (AI) studies using radiography for TMJ OA diagnosis reported a pooled sensitivity of 0.76 and specificity of 0.79, based on a total of 10,077 TMJ images from seven studies. While three of these studies focused on panoramic radiography and various transfer learning models, others investigated the 3D shape of the condyle and disease classification using CBCT images 5 . Most existing AI research on TMJ OA diagnosis has relied primarily on single imaging modalities with little incorporation of clinical data. Although the value of joint noise data in TMD diagnosis is not always definitive, some studies have underscored its importance. For instance, click sounds have been found to be more strongly associated with disk displacement without reduction 6 , and coarse crepitus has been identified as a useful clinical indicator for TMJ osteoarthritis 7 . The Diagnostic Criteria for Temporomandibular Disorders (DC/TMD) also recommends utilizing all joint noise data in diagnosing degenerative joint disease (DJD) 8 . This study aimed to develop and evaluate an AI model that integrates TMJ panoramic radiography and joint noise data to improve the screening and diagnosis of TMJ OA. By aligning with current diagnostic criteria and addressing the limitations of previous AI studies that relied solely on imaging data, this approach offers the potential for a more comprehensive and accurate diagnostic tool for TMJ OA. Results Among the six AI models evaluated, GoogleNet consistently demonstrated superior performance, as evidenced by the highest F1-score across all conditions, regardless of the presence or type of joint noise data (Table 1 ). In the assessment of models incorporating various joint noise data inputs using GoogleNet, all experiments were conducted in triplicate to ensure reliability, with mean values utilized for analysis. The integration of crepitus data resulted in an overall performance enhancement (Model 1 vs. 2). Conversely, the inclusion of subjective joint noise data exhibited a stabilizing effect on label-specific predictions (Model 1 vs. 4). Model 2 and 4 achieved the highest F1-score (0.72) (Table 2 , Fig. 1). Table 1 Comparison of F1-scores among various pre-trained models for TMJ OA diagnosis. TMJ OA, temporomandibular joint osteoarthritis. Model Res18 Res50 Res101 VGG16 VGG19 GoogleNet 1 0.67 0.69 0.68 0.69 0.69 0.70 2 0.71 0.71 0.70 0.71 0.71 0.72 5 0.69 0.68 0.64 0.67 0.69 0.71 6 0.70 0.70 0.66 0.71 0.69 0.72 Table 2 Detailed performance metrics for different AI models in TMJ OA diagnosis. AI, artificial intelligence; TMJ OA, temporomandibular joint osteoarthritis; D, osteoarthritis; N, normal. Model Precision Recall F1-score Accuracy Macro average F1-score 1 True = D 0.78 0.53 0.63 0.72 0.70 True = N 0.69 0.88 0.77 2 True = D 0.81 0.55 0.66 0.74 0.72 True = N 0.71 0.89 0.79 3 True = D 0.69 0.63 0.66 0.70 0.70 True = N 0.71 0.76 0.74 4 True = D 0.71 0.67 0.69 0.72 0.72 True = N 0.74 0.77 0.75 5 True = D 0.69 0.61 0.65 0.69 0.69 True = N 0.70 0.77 0.73 6 True = D 0.74 0.58 0.65 0.72 0.71 True = N 0.71 0.83 0.76 In selecting the best model, we focused on TMJ OA (D) recall (sensitivity) and normal (N) precision (positive predictive value) as key metrics, given their importance in medical diagnostics, especially for screening tools. D recall represents the proportion of actual OA cases correctly identified by the model. High D recall is crucial for a screening tool because it minimizes false negatives, ensuring that fewer cases of TMJ OA are missed. This is particularly important in early detection and intervention of TMJ OA. N precision, on the other hand, indicates the model's accuracy in classifying normal cases. High N precision means that when the model classifies a case as normal, there's a high probability that it is indeed normal. This is important for reducing unnecessary further testing or treatment for patients who don't have TMJ OA. By prioritizing these metrics, we aimed to develop a screening tool that effectively identifies potential TMJ OA cases (high D recall) while also accurately ruling out normal cases (high N precision). This approach helps to balance the need for early detection with the goal of minimizing unnecessary interventions or further diagnostic procedures. The ranking of model suitability, calculated by summing the ranks of macro average F1, OA recall, and N precision, yielded the following order: 4, 2, 6, 1, 3, 5. To identify the image areas that the AI model relied on for its judgment, we utilized a saliency map, as shown in Fig. 2. Comparative analysis between the AI models and orofacial pain specialists was shown in Table 3 and Fig. 3. AI models outperformed specialists across all metrics (precision, recall, F1-score) for both Model 1 (image-only) and Model 4 (image and whole joint noise data). The AI's diagnostic capability improved with the addition of joint noise data (Model 4), as evidenced by the increase in macro average F1-score from 0.70 (Model 1) to 0.72 (Model 4). Conversely, specialists showed a decline in performance when presented with both image and joint noise data (Model 4) compared to image-only assessment (Model 1), with macro average F1-scores decreasing from 0.63 to 0.59. Table 3 Average diagnostic performance of eight orofacial pain specialists in TMJ OA diagnosis. TMJ OA, temporomandibular joint osteoarthritis; D, osteoarthritis; N, normal. Model Precision Recall F1-score Accuracy Macro average F1-score 1 True = D 0.69 0.50 0.58 0.64 0.63 True = N 0.61 0.77 0.68 4 True = D 0.65 0.44 0.52 0.60 0.59 True = N 0.57 0.76 0.66 Discussion Our study reveals important findings regarding the diagnosis of TMJ OA using AI models and the significance of various diagnostic factors, particularly joint noise. The most notable finding is that combining both subjective and objective noise data (Model 4) yielded the highest performance, emphasizing the importance of using all available auditory information. This aligns with the DC/TMD for diagnosing DJD, which now includes both fine and coarse crepitus, as well as patient-reported joint noises 8 . However, for DJD diagnosis, imaging such as CBCT remains essential due to the low sensitivity (0.55) and specificity (0.61) when relying solely on clinical exams without imaging. While integrating all types of noise data proved most effective, using only crepitus also enhanced model performance. Crepitus, a well-established marker in TMJ OA diagnostics, has shown high specificity in previous studies 9 . Our findings corroborate this, as models incorporating crepitus (Models 2, 4, and 6) outperformed those relying solely on image data (Model 1). Interestingly, while our AI models, particularly those integrating both imaging and noise data, outperformed orofacial pain specialists, it is important to consider the context of this comparison. Specialists typically rely on a broader range of diagnostic information during clinical evaluations, including patient history, mandibular movements, muscle and joint palpation, joint sounds, and occlusion, which were not available in this study. The AI models, on the other hand, were limited to analyzing only the panoramic images and noise data. This limitation may explain why the AI models performed better under these conditions. In diagnosing TMJ OA, research findings are mixed regarding the effectiveness of TMJ panoramic radiography compared to panoramic radiography. TMJ panoramic radiography has been shown to have limited but improved diagnostic accuracy over conventional panoramic radiography, particularly in detecting bony lesions like flattening, erosion, and osteophytes on the mandibular condyle 10 . The combination of lateral and frontal TMJ projections often shows the highest sensitivity for detecting these lesions, though with varying levels of specificity and overall accuracy. In contrast, other studies have reported that general panoramic radiography tends to have better diagnostic accuracy than TMJ panoramic projections, especially when evaluating condylar cortical erosion 11 . The broader context provided by general panoramic radiography may aid in more precise assessments. Despite the promising results from our AI models, their performance was lower than expected compared to studies using general panoramic radiography for TMJ OA diagnosis. However, our study focused on detecting deformation due to subcortical cysts, surface erosion, osteophytes, and generalized sclerosis, which may account for some differences in findings. Also, the challenges associated with TMJ panoramic radiography—such as movement artifacts due to open-mouth positioning and the potential for obscuring key joint structures—likely contributed to the reduced effectiveness observed in our results. Despite the promising results, this study has several limitations that should be considered for future research. Although our dataset was labeled based on CBCT interpretations by multiple dental radiology specialists, relying on the CBCT images themselves for labeling might improve the precision and consistency of the AI models. Additionally, the incomplete collection of subjective noise data, as not all cases included this information, may have limited the accuracy and generalizability of our findings. Finally, the study's limited sample size and single-center design suggest the need for further research that includes more comprehensive datasets and multi-center data to validate and refine the AI models for broader clinical application. In conclusion, our study demonstrates that AI-enhanced TMJ OA diagnosis, particularly when combining panoramic radiography with joint noise data, offers a promising approach for early detection and improved patient care. However, further research is essential to fully optimize this approach, especially considering the evolving diagnostic criteria and the specific challenges of TMJ imaging. Methods Study design and data collection This study was approved by the Institutional Review Board (IRB) of Yonsei University College of Dentistry (IRB 2-2024-0011), and all methods were performed in accordance with relevant guidelines and regulations. The IRB waived the requirement for informed consent documentation. Radiographic images were reviewed from patients who visited the Orofacial Pain Clinic at Yonsei University Dental Hospital between January 2019 and February 2021, reported TMD-related symptoms, and had both TMJ panoramic radiography (Rayscan Alpha Plus, Ray Co. Ltd., Hwaseong-si, Korea) and TMJ CBCT (Alphard 3030 device, Asahi Roentgen Ind., Co. Ltd, Kyoto, Japan) evaluated by oral and maxillofacial radiology specialists. Exclusion criteria included patients under 18 years of age, those with a history of orthognathic surgery, macro trauma, systemic diseases causing joint deformity, or a time gap of more than three months between the TMJ panorama and CBCT imaging. For the collection of joint noise data, both objective (clinician-detected crepitus) and subjective noise reported by patients were reviewed retrospectively through electronic medical records (EMR). Crepitus was defined as a continuous or multiple coarse friction sound/sensation felt by the examiner's finger placed anterior to the external auditory meatus during three repetitions of opening and closing movements, as well as during lateral and protrusive movements of the mandible. This objective assessment was performed by trained clinicians as part of the standard TMJ examination protocol. Patient-reported subjective sounds were categorized into four types: P: clicking or popping sound R: crepitus or grating sound like gravel or sand moving PR: both joint noises present I: unspecified sound (patient reports a sound but cannot accurately describe its nature) Data Processing A total of 2,631 TMJ panoramic images were collected for this study. While there was only a slight difference in AI model performance between open and closed TMJ panoramic images, with the open images showing a 2–3% higher performance, this study utilized only open ones (Supplemental Table). Each image was re-labeled by an orofacial pain specialist as either "normal (N)" or "TMJ OA (D)" based on corresponding CBCT analysis criteria for temporomandibular disorder diagnosis 8 , 12 . Following the exclusion of 1,354 indeterminate OA images and errors 13 , the final dataset included 2,127 images labeled as "N" and 1,781 images labeled as "D". Crepitus data was collected for the entire dataset, while subjective noise data was collected for a subset of samples (48.13% of the total). AI model development The AI model employed pre-trained CNN architectures to extract feature vectors, which were then combined with joint noise information for final diagnosis. We compared the performance of several pre-trained models commonly used for image classification tasks, including ResNet, VGGNet, and GoogleNet. Data preprocessing involved rescaling (to 1/4, 1/2, and full original size). To enhance the robustness of the prediction model, data augmentation techniques were applied, including horizontal flipping, sharpness adjustment, brightness modification, and random rotation. To evaluate the effectiveness of sound information in the AI model, we incorporated joint noise information into the feature vector. We recorded objective noise (clinician-detected crepitus) and subjective noise reported by patients for each image. While crepitus data was collected for the entire dataset, subjective noise data was collected for 48.13% of the samples. Both crepitus and three types of patient-reported subjective sounds were binarized and encoded as four-dimensional vectors, with 1 indicating the presence of a sound and 0 indicating its absence. The integration of feature vectors and sound information was achieved by merging binarized sound information with the feature vectors derived from the TMJ panoramic images through the pre-trained models. A single-layer neural network used the merged vectors to perform the final diagnosis. To improve the performance of the AI model, experiments were conducted on six models by varying their inputs. The six models were: Model 1: image only Model 2: image and crepitus Model 3: image and subjective sounds (P, R, PR, I) Model 4: image, crepitus and subjective sounds (P, R, PR, I) Model 5: image and certain subjective sounds (R, PR) Model 6: image, crepitus and certain subjective sounds (R, PR) The structure of integrated AI model architecture was shown in Fig. 4 Evaluation of AI models' performance and clinical usability The diagnostic performance of the AI model was evaluated by splitting the dataset into 60% training data and 40% test data. We conducted three repetitions of experiments with different training/test samplings. Due to class imbalance in the dataset, we used the F1-score instead of accuracy for performance evaluation. To train the AI model, we used the stochastic gradient descent (SGD) optimizer with a learning rate of 1.0 × 10 –3 . We trained AI models for a total of 300 epochs. The design, training, and evaluation of the diagnostic model were implemented using Python, with the PyTorch library utilized for designing and training the deep learning models. The saliency map was used to interpret and visualize the regions of the image considered by the AI while making predictions. For the clinical usability evaluation, 100 images labeled as "D" and 100 images labeled as "N" were randomly selected and mixed. These images were then reviewed by eight orofacial pain specialists. Each specialist evaluated the images twice, with and without noise joint information, at one-week intervals. Declarations Author contributions E.C. contributed to the conception, design, data acquisition, analysis, and interpretation, drafted and critically revised the manuscript. Y.S. contributed to the design, data analysis, and interpretation, and critically revised the manuscript. S.T.K. contributed to the conception, design, critically revised the manuscript. S.S., K.L. and T.A. contributed to data analysis, interpretation, and original draft. Y.S. contributed to the design, data analysis, and interpretation, and critically revised the manuscript. R.L. and S.K. contributed to data acquisition. All authors gave final approval and agreed to be accountable for all aspects of the work. Data availability The data that support the findings of this study are available from the corresponding author upon reasonable request. Funding This study was supported by the Yonsei University College of Dentistry Fund (6-2024-0005) and the National Research Foundation of Korea (NRF) grant funded by the Ministry of Science and ICT (MSIT) of Korea (RS-2023-00208412). Competing interests The authors declare no competing interests. References Wang, X., Zhang, J., Gan, Y. & Zhou, Y. Current understanding of pathogenesis and treatment of TMJ osteoarthritis. J. Dent. Res. 94 , 666–673 (2015). Song, H., Lee, J. Y., Huh, K. H. & Park, J. W. Long-term changes of temporomandibular joint osteoarthritis on computed tomography. Sci. Rep. 10 , 6731. https://doi.org/10.1038/s41598-020-63493-8 (2020). Kim, K., Wojczyńska, A. & Lee, J. Y. The incidence of osteoarthritic change on computed tomography of Korean temporomandibular disorder patients diagnosed by RDC/TMD; a retrospective study. Acta Odontol. Scand. 74 , 337–342. https://doi.org/10.3109/00016357.2015.1136678 (2016). Schiffman, E. L. et al. The research diagnostic criteria for temporomandibular disorders. I: overview and methodology for assessment of validity. J. Orofac. Pain . 24 , 7 (2010). Almășan, O., Leucuța, D. C., Hedeșiu, M., Mureșanu, S. & Popa, Ș. L. Temporomandibular joint osteoarthritis diagnosis employing artificial intelligence: systematic review and meta-analysis. J. Clin. Med. 12 , 942. https://doi.org/10.3390/jcm12030942 (2023). Manfredini, D., Basso, D., Salmaso, L. & Guarda-Nardini, L. Temporomandibular joint click sound and magnetic resonance-depicted disk position: which relationship? J. Dent. 36 , 256–260. https://doi.org/10.1016/j.jdent.2008.01.002 (2008). Wiese, M. et al. Association between temporomandibular joint symptoms, signs, and clinical diagnosis using the RDC/TMD and radiographic findings in temporomandibular joint tomograms. J. Orofac. Pain . 22 , 239–251 (2008). Schiffman, E. et al. Diagnostic criteria for temporomandibular disorders (DC/TMD) for clinical and research applications: recommendations of the international RDC/TMD consortium network and orofacial pain special interest group. J. Oral Facial Pain Headache . 28 , 6–27. https://doi.org/10.11607/jop.1151 (2014). Abrahamsson, A. K. et al. Frequency of temporomandibular joint osteoarthritis and related symptoms in a hand osteoarthritis cohort. Osteoarthr. Cartil. 25 , 654–657 (2017). Im, Y. G. et al. Diagnostic accuracy and reliability of panoramic temporomandibular joint (TMJ) radiography to detect bony lesions in patients with TMJ osteoarthritis. J. Dent. Sci. 13 , 396–404 (2018). Honey, O. B. et al. Accuracy of cone-beam computed tomography imaging of the temporomandibular joint: comparisons with panoramic radiology and linear tomography. Am. J. Orthod. Dentofac. Orthop. 132 , 429–438. https://doi.org/10.1016/j.ajodo.2005.10.032 (2007). Ahmad, M. et al. Research diagnostic criteria for temporomandibular disorders (RDC/TMD): development of image analysis criteria and examiner reliability for image analysis. Oral Surg. Oral Med. Oral Pathol. Oral Radiol. Endod . 107 , 844–860 (2009). Choi, E., Kim, D., Lee, J. Y. & Park, H. K. Artificial intelligence in detecting temporomandibular joint osteoarthritis on orthopantomogram. Sci. Rep. 11 , 10246. https://doi.org/10.1038/s41598-021-89742-y (2021). Additional Declarations No competing interests reported. Supplementary Files Supp0913.docx Cite Share Download PDF Status: Published Journal Publication published 13 Jan, 2025 Read the published version in Scientific Reports → Version 1 posted Editorial decision: Revision requested 07 Nov, 2024 Reviews received at journal 27 Oct, 2024 Reviewers agreed at journal 25 Oct, 2024 Reviewers agreed at journal 24 Oct, 2024 Reviews received at journal 11 Oct, 2024 Reviewers agreed at journal 29 Sep, 2024 Reviewers invited by journal 28 Sep, 2024 Editor assigned by journal 28 Sep, 2024 Editor invited by journal 25 Sep, 2024 Submission checks completed at journal 25 Sep, 2024 First submitted to journal 13 Sep, 2024 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-5086242","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":375151219,"identity":"ddac058f-9280-4481-aa24-07b174b63dd4","order_by":0,"name":"Eunhye Choi","email":"","orcid":"","institution":"Seoul National University","correspondingAuthor":false,"prefix":"","firstName":"Eunhye","middleName":"","lastName":"Choi","suffix":""},{"id":375151220,"identity":"6adc4e7e-283b-4622-b40a-c4df1c60869c","order_by":1,"name":"Seokwon Shin","email":"","orcid":"","institution":"Dongguk University-Seoul","correspondingAuthor":false,"prefix":"","firstName":"Seokwon","middleName":"","lastName":"Shin","suffix":""},{"id":375151221,"identity":"afc12cb3-991f-4fff-927c-a394e933609e","order_by":2,"name":"Kijin Lee","email":"","orcid":"","institution":"Dongguk University-Seoul","correspondingAuthor":false,"prefix":"","firstName":"Kijin","middleName":"","lastName":"Lee","suffix":""},{"id":375151222,"identity":"c761253c-a4bb-4376-92aa-ba37bf0e64fa","order_by":3,"name":"Taejin An","email":"","orcid":"","institution":"Dongguk University-Seoul","correspondingAuthor":false,"prefix":"","firstName":"Taejin","middleName":"","lastName":"An","suffix":""},{"id":375151223,"identity":"6946bac0-beb3-43a0-a78d-db5e86b8cf42","order_by":4,"name":"Richard K. Lee","email":"","orcid":"","institution":"University of Technology Sydney","correspondingAuthor":false,"prefix":"","firstName":"Richard","middleName":"K.","lastName":"Lee","suffix":""},{"id":375151224,"identity":"8ad5181e-c58d-4248-adac-e13709adfdc3","order_by":5,"name":"Sunmin Kim","email":"","orcid":"","institution":"Boston University","correspondingAuthor":false,"prefix":"","firstName":"Sunmin","middleName":"","lastName":"Kim","suffix":""},{"id":375151225,"identity":"67b1b111-23d3-4b57-9c5a-fe080df8ddff","order_by":6,"name":"Youngdoo Son","email":"","orcid":"","institution":"Dongguk University-Seoul","correspondingAuthor":false,"prefix":"","firstName":"Youngdoo","middleName":"","lastName":"Son","suffix":""},{"id":375151226,"identity":"fd86bff0-fa3a-4a39-bf12-7299f428e804","order_by":7,"name":"Seong Teak Kim","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA20lEQVRIie3QMQuCQBTA8SeCNVzYeHJQX+Hk4Fqkz1IITi4tITR0ky1Cq30OwfnCwcUP4NBQBM1uOUU5NZ5uQfef3w/eewA63Q9Gy+OzaegL2QBWP+ImkhtpJGeO6EsYXnFzXElGZV/CkQzuRnxZZ3XIby0s50rijUTJNvFjndfhgiHwXaEk6CzIKTY7wgmAuVIvhn0gkw/J0pA7LezVhOHAIqgqGMUhxwgKNXGTynTTKJjh6rEliJbqW2iZGNeGesg++LnTRjv1x75Nu5XoAABgy0HjOp1O90e9ATDDPjPHPuwHAAAAAElFTkSuQmCC","orcid":"","institution":"Yonsei University College of Dentistry","correspondingAuthor":true,"prefix":"","firstName":"Seong","middleName":"Teak","lastName":"Kim","suffix":""}],"badges":[],"createdAt":"2024-09-14 00:24:10","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-5086242/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-5086242/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1038/s41598-024-83750-4","type":"published","date":"2025-01-13T15:57:22+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":69962417,"identity":"fc4064fb-ffa5-4c36-b1a7-dd6cda39ecde","added_by":"auto","created_at":"2024-11-27 04:51:26","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":22013,"visible":true,"origin":"","legend":"\u003cp\u003ePerformance comparison of six AI models in diagnosing TMJ OA. AI, artificial intelligence; TMJ OA, temporomandibular joint osteoarthritis; D, osteoarthritis; N, normal.\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-5086242/v1/a57ee14f0dc636bb7fbec31a.png"},{"id":69962418,"identity":"529df3c9-eb3a-4382-8f07-f0701877ad26","added_by":"auto","created_at":"2024-11-27 04:51:26","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":646456,"visible":true,"origin":"","legend":"\u003cp\u003eSaliency map highlighting the region of interest in AI-based TMJ OA diagnosis. AI, artificial intelligence; TMJ OA, temporomandibular joint osteoarthritis.\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-5086242/v1/58dac078bc2bec879f6d29f9.png"},{"id":69962419,"identity":"a04599bf-5c2e-4222-94e6-63b9313232f5","added_by":"auto","created_at":"2024-11-27 04:51:27","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":35348,"visible":true,"origin":"","legend":"\u003cp\u003eDiagnostic performance comparison: specialists vs. AI models. AI, artificial intelligence.\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-5086242/v1/7211fa330c65598c7a38bd63.png"},{"id":69963641,"identity":"a9b9e726-86fd-44e1-b250-a95dd75088fa","added_by":"auto","created_at":"2024-11-27 04:59:27","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":185989,"visible":true,"origin":"","legend":"\u003cp\u003eArchitecture of the integrated AI model for TMJ OA diagnosis. AI, artificial intelligence; TMJ OA, temporomandibular joint osteoarthritis.\u003c/p\u003e","description":"","filename":"4.png","url":"https://assets-eu.researchsquare.com/files/rs-5086242/v1/15f32b763c006929692f8a9e.png"},{"id":74284749,"identity":"0883d20c-14d6-4c35-a122-6e1c5c7ba834","added_by":"auto","created_at":"2025-01-20 16:12:11","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1733969,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-5086242/v1/ac0cd440-e23b-4229-b875-6d98adbad88f.pdf"},{"id":69962420,"identity":"79b4ae3c-4a1f-4be7-8051-b8c0e0b2e462","added_by":"auto","created_at":"2024-11-27 04:51:27","extension":"docx","order_by":3,"title":"","display":"","copyAsset":false,"role":"supplement","size":14555,"visible":true,"origin":"","legend":"","description":"","filename":"Supp0913.docx","url":"https://assets-eu.researchsquare.com/files/rs-5086242/v1/b5ac2bc21942ccb014b0333f.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"Artificial intelligence-enhanced diagnosis of temporomandibular joint osteoarthritis using temporomandibular joint panoramic radiography and joint noise data","fulltext":[{"header":"Introduction","content":"\u003cp\u003eTemporomandibular joint osteoarthritis (TMJ OA) is a degenerative condition characterized by cartilage breakdown and bone deformation in the mandibular condyle and articular fossa, significantly affecting patients\u0026apos; quality of life by impairing function and aesthetics\u003csup\u003e1\u003c/sup\u003e. Interestingly, while osteoarthritis (OA) is typically considered female predominant and age-related, the linear correlation between bone change and age observed in other joints is not evident in the temporomandibular joint (TMJ). Temporomandibular disorder (TMD) is most prevalent in young adults aged 20 to 40, and TMJ OA may begin at an early age\u003csup\u003e2\u003c/sup\u003e. Notably, the severity of TMJ OA does not always correspond with pain intensity; TMJ OA changes in young Korean TMD patients are as common as in older individuals, showing no correlation with clinical pain or disc displacement\u003csup\u003e3\u003c/sup\u003e. This makes early detection both challenging and essential, especially since significant degeneration often occurs without overt symptoms.\u003c/p\u003e\n\u003cp\u003eCone-beam computed tomography (CBCT) is the reference standard for diagnosing TMJ OA due to its high diagnostic accuracy and detailed anatomical assessment\u003csup\u003e4\u003c/sup\u003e. However, CBCT is costly and exposes patients to higher radiation doses compared to other imaging modalities. While panoramic radiography is more accessible and cost-effective, it lacks comprehensive evaluation capabilities for the TMJ region. Conversely, TMJ panoramic radiography provides a more targeted evaluation, enabling assessment of the mandibular condyle\u0026apos;s range of motion without significant projection limitations.\u003cbr\u003eRecent a systematic review and meta-analysis of artificial intelligence (AI) studies using radiography for TMJ OA diagnosis reported a pooled sensitivity of 0.76 and specificity of 0.79, based on a total of 10,077 TMJ images from seven studies. While three of these studies focused on panoramic radiography and various transfer learning models, others investigated the 3D shape of the condyle and disease classification using CBCT images\u003csup\u003e5\u003c/sup\u003e.\u0026nbsp;Most existing AI research on TMJ OA diagnosis has relied primarily on single imaging modalities with little incorporation of clinical data. Although the value of joint noise data in TMD diagnosis is not always definitive, some studies have underscored its importance. For instance, click sounds have been found to be more strongly associated with disk displacement without reduction\u003csup\u003e6\u003c/sup\u003e, and coarse crepitus has been identified as a useful clinical indicator for TMJ osteoarthritis\u003csup\u003e7\u003c/sup\u003e. The Diagnostic Criteria for Temporomandibular Disorders (DC/TMD) also recommends utilizing all joint noise data in diagnosing degenerative joint disease (DJD)\u003csup\u003e8\u003c/sup\u003e.\u003c/p\u003e\n\u003cp\u003eThis study aimed to develop and evaluate an AI model that integrates TMJ panoramic radiography and joint noise data to improve the screening and diagnosis of TMJ OA. By aligning with current diagnostic criteria and addressing the limitations of previous AI studies that relied solely on imaging data, this approach offers the potential for a more comprehensive and accurate diagnostic tool for TMJ OA.\u003c/p\u003e"},{"header":"Results","content":"\u003cp\u003eAmong the six AI models evaluated, GoogleNet consistently demonstrated superior performance, as evidenced by the highest F1-score across all conditions, regardless of the presence or type of joint noise data (Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e). In the assessment of models incorporating various joint noise data inputs using GoogleNet, all experiments were conducted in triplicate to ensure reliability, with mean values utilized for analysis. The integration of crepitus data resulted in an overall performance enhancement (Model 1 vs. 2). Conversely, the inclusion of subjective joint noise data exhibited a stabilizing effect on label-specific predictions (Model 1 vs. 4). Model 2 and 4 achieved the highest F1-score (0.72) (Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e, Fig.\u0026nbsp;1).\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\u003eComparison of F1-scores among various pre-trained models for TMJ OA diagnosis. TMJ OA, temporomandibular joint osteoarthritis.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"7\"\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 \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 \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eModel\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eRes18\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eRes50\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eRes101\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eVGG16\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eVGG19\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e \u003cp\u003eGoogleNet\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.67\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.69\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.68\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.69\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.69\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.70\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.71\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.71\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.70\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.71\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.71\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.72\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.69\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.68\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.64\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.67\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.69\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.71\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e6\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.70\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.70\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.66\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.71\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.69\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.72\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\u003eDetailed performance metrics for different AI models in TMJ OA diagnosis. AI, artificial intelligence; TMJ OA, temporomandibular joint osteoarthritis; D, osteoarthritis; N, normal.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"7\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" 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 \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 \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eModel\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003ePrecision\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eRecall\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eF1-score\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eAccuracy\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e \u003cp\u003eMacro average\u003c/p\u003e \u003cp\u003eF1-score\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003e1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTrue\u0026thinsp;=\u0026thinsp;D\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.78\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.53\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.63\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003e0.72\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003e0.70\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTrue\u0026thinsp;=\u0026thinsp;N\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.69\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.88\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.77\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003e2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTrue\u0026thinsp;=\u0026thinsp;D\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.81\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.55\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.66\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003e0.74\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003e0.72\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTrue\u0026thinsp;=\u0026thinsp;N\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.71\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.89\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.79\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003e3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTrue\u0026thinsp;=\u0026thinsp;D\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.69\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.63\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.66\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003e0.70\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003e0.70\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTrue\u0026thinsp;=\u0026thinsp;N\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.71\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.76\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.74\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003e4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTrue\u0026thinsp;=\u0026thinsp;D\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.71\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.67\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.69\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003e0.72\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003e0.72\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTrue\u0026thinsp;=\u0026thinsp;N\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.74\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.77\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.75\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003e5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTrue\u0026thinsp;=\u0026thinsp;D\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.69\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.61\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.65\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003e0.69\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003e0.69\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTrue\u0026thinsp;=\u0026thinsp;N\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.70\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.77\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.73\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003e6\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTrue\u0026thinsp;=\u0026thinsp;D\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.74\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.58\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.65\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003e0.72\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003e0.71\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTrue\u0026thinsp;=\u0026thinsp;N\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.71\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.83\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.76\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\u003eIn selecting the best model, we focused on TMJ OA (D) recall (sensitivity) and normal (N) precision (positive predictive value) as key metrics, given their importance in medical diagnostics, especially for screening tools. D recall represents the proportion of actual OA cases correctly identified by the model. High D recall is crucial for a screening tool because it minimizes false negatives, ensuring that fewer cases of TMJ OA are missed. This is particularly important in early detection and intervention of TMJ OA. N precision, on the other hand, indicates the model's accuracy in classifying normal cases. High N precision means that when the model classifies a case as normal, there's a high probability that it is indeed normal. This is important for reducing unnecessary further testing or treatment for patients who don't have TMJ OA. By prioritizing these metrics, we aimed to develop a screening tool that effectively identifies potential TMJ OA cases (high D recall) while also accurately ruling out normal cases (high N precision). This approach helps to balance the need for early detection with the goal of minimizing unnecessary interventions or further diagnostic procedures. The ranking of model suitability, calculated by summing the ranks of macro average F1, OA recall, and N precision, yielded the following order: 4, 2, 6, 1, 3, 5.\u003c/p\u003e \u003cp\u003eTo identify the image areas that the AI model relied on for its judgment, we utilized a saliency map, as shown in Fig.\u0026nbsp;2.\u003c/p\u003e \u003cp\u003eComparative analysis between the AI models and orofacial pain specialists was shown in Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e and Fig.\u0026nbsp;3. AI models outperformed specialists across all metrics (precision, recall, F1-score) for both Model 1 (image-only) and Model 4 (image and whole joint noise data). The AI's diagnostic capability improved with the addition of joint noise data (Model 4), as evidenced by the increase in macro average F1-score from 0.70 (Model 1) to 0.72 (Model 4). Conversely, specialists showed a decline in performance when presented with both image and joint noise data (Model 4) compared to image-only assessment (Model 1), with macro average F1-scores decreasing from 0.63 to 0.59.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab3\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eAverage diagnostic performance of eight orofacial pain specialists in TMJ OA diagnosis. TMJ OA, temporomandibular joint osteoarthritis; D, osteoarthritis; N, normal.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"7\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" 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 \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 \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eModel\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003ePrecision\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eRecall\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eF1-score\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eAccuracy\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e \u003cp\u003eMacro average\u003c/p\u003e \u003cp\u003eF1-score\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003e1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTrue\u0026thinsp;=\u0026thinsp;D\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.69\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.50\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.58\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003e0.64\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003e0.63\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTrue\u0026thinsp;=\u0026thinsp;N\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.61\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.77\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.68\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003e4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTrue\u0026thinsp;=\u0026thinsp;D\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.65\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.44\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.52\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003e0.60\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003e0.59\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTrue\u0026thinsp;=\u0026thinsp;N\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.57\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.76\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.66\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eOur study reveals important findings regarding the diagnosis of TMJ OA using AI models and the significance of various diagnostic factors, particularly joint noise.\u003c/p\u003e \u003cp\u003eThe most notable finding is that combining both subjective and objective noise data (Model 4) yielded the highest performance, emphasizing the importance of using all available auditory information. This aligns with the DC/TMD for diagnosing DJD, which now includes both fine and coarse crepitus, as well as patient-reported joint noises\u003csup\u003e\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e\u003c/sup\u003e. However, for DJD diagnosis, imaging such as CBCT remains essential due to the low sensitivity (0.55) and specificity (0.61) when relying solely on clinical exams without imaging.\u003c/p\u003e \u003cp\u003eWhile integrating all types of noise data proved most effective, using only crepitus also enhanced model performance. Crepitus, a well-established marker in TMJ OA diagnostics, has shown high specificity in previous studies\u003csup\u003e\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e\u003c/sup\u003e. Our findings corroborate this, as models incorporating crepitus (Models 2, 4, and 6) outperformed those relying solely on image data (Model 1).\u003c/p\u003e \u003cp\u003eInterestingly, while our AI models, particularly those integrating both imaging and noise data, outperformed orofacial pain specialists, it is important to consider the context of this comparison. Specialists typically rely on a broader range of diagnostic information during clinical evaluations, including patient history, mandibular movements, muscle and joint palpation, joint sounds, and occlusion, which were not available in this study. The AI models, on the other hand, were limited to analyzing only the panoramic images and noise data. This limitation may explain why the AI models performed better under these conditions.\u003c/p\u003e \u003cp\u003eIn diagnosing TMJ OA, research findings are mixed regarding the effectiveness of TMJ panoramic radiography compared to panoramic radiography. TMJ panoramic radiography has been shown to have limited but improved diagnostic accuracy over conventional panoramic radiography, particularly in detecting bony lesions like flattening, erosion, and osteophytes on the mandibular condyle\u003csup\u003e\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e\u003c/sup\u003e. The combination of lateral and frontal TMJ projections often shows the highest sensitivity for detecting these lesions, though with varying levels of specificity and overall accuracy. In contrast, other studies have reported that general panoramic radiography tends to have better diagnostic accuracy than TMJ panoramic projections, especially when evaluating condylar cortical erosion\u003csup\u003e\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e\u003c/sup\u003e. The broader context provided by general panoramic radiography may aid in more precise assessments. Despite the promising results from our AI models, their performance was lower than expected compared to studies using general panoramic radiography for TMJ OA diagnosis. However, our study focused on detecting deformation due to subcortical cysts, surface erosion, osteophytes, and generalized sclerosis, which may account for some differences in findings. Also, the challenges associated with TMJ panoramic radiography\u0026mdash;such as movement artifacts due to open-mouth positioning and the potential for obscuring key joint structures\u0026mdash;likely contributed to the reduced effectiveness observed in our results.\u003c/p\u003e \u003cp\u003eDespite the promising results, this study has several limitations that should be considered for future research. Although our dataset was labeled based on CBCT interpretations by multiple dental radiology specialists, relying on the CBCT images themselves for labeling might improve the precision and consistency of the AI models. Additionally, the incomplete collection of subjective noise data, as not all cases included this information, may have limited the accuracy and generalizability of our findings. Finally, the study's limited sample size and single-center design suggest the need for further research that includes more comprehensive datasets and multi-center data to validate and refine the AI models for broader clinical application.\u003c/p\u003e \u003cp\u003eIn conclusion, our study demonstrates that AI-enhanced TMJ OA diagnosis, particularly when combining panoramic radiography with joint noise data, offers a promising approach for early detection and improved patient care. However, further research is essential to fully optimize this approach, especially considering the evolving diagnostic criteria and the specific challenges of TMJ imaging.\u003c/p\u003e"},{"header":"Methods","content":"\u003cdiv id=\"Sec4\" class=\"Section2\"\u003e \u003ch2\u003eStudy design and data collection\u003c/h2\u003e \u003cp\u003e This study was approved by the Institutional Review Board (IRB) of Yonsei University College of Dentistry (IRB 2-2024-0011), and all methods were performed in accordance with relevant guidelines and regulations. The IRB waived the requirement for informed consent documentation.\u003c/p\u003e \u003cp\u003eRadiographic images were reviewed from patients who visited the Orofacial Pain Clinic at Yonsei University Dental Hospital between January 2019 and February 2021, reported TMD-related symptoms, and had both TMJ panoramic radiography (Rayscan Alpha Plus, Ray Co. Ltd., Hwaseong-si, Korea) and TMJ CBCT (Alphard 3030 device, Asahi Roentgen Ind., Co. Ltd, Kyoto, Japan) evaluated by oral and maxillofacial radiology specialists. Exclusion criteria included patients under 18 years of age, those with a history of orthognathic surgery, macro trauma, systemic diseases causing joint deformity, or a time gap of more than three months between the TMJ panorama and CBCT imaging.\u003c/p\u003e \u003cp\u003eFor the collection of joint noise data, both objective (clinician-detected crepitus) and subjective noise reported by patients were reviewed retrospectively through electronic medical records (EMR). Crepitus was defined as a continuous or multiple coarse friction sound/sensation felt by the examiner's finger placed anterior to the external auditory meatus during three repetitions of opening and closing movements, as well as during lateral and protrusive movements of the mandible. This objective assessment was performed by trained clinicians as part of the standard TMJ examination protocol. Patient-reported subjective sounds were categorized into four types:\u003c/p\u003e \u003cp\u003e \u003col\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eP: clicking or popping sound\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eR: crepitus or grating sound like gravel or sand moving\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003ePR: both joint noises present\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eI: unspecified sound (patient reports a sound but cannot accurately describe its nature)\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003c/ol\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec5\" class=\"Section2\"\u003e \u003ch2\u003eData Processing\u003c/h2\u003e \u003cp\u003eA total of 2,631 TMJ panoramic images were collected for this study. While there was only a slight difference in AI model performance between open and closed TMJ panoramic images, with the open images showing a 2\u0026ndash;3% higher performance, this study utilized only open ones (Supplemental Table). Each image was re-labeled by an orofacial pain specialist as either \"normal (N)\" or \"TMJ OA (D)\" based on corresponding CBCT analysis criteria for temporomandibular disorder diagnosis\u003csup\u003e\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e,\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e\u003c/sup\u003e. Following the exclusion of 1,354 indeterminate OA images and errors\u003csup\u003e\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e\u003c/sup\u003e, the final dataset included 2,127 images labeled as \"N\" and 1,781 images labeled as \"D\".\u003c/p\u003e \u003cp\u003eCrepitus data was collected for the entire dataset, while subjective noise data was collected for a subset of samples (48.13% of the total).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec6\" class=\"Section2\"\u003e \u003ch2\u003eAI model development\u003c/h2\u003e \u003cp\u003eThe AI model employed pre-trained CNN architectures to extract feature vectors, which were then combined with joint noise information for final diagnosis. We compared the performance of several pre-trained models commonly used for image classification tasks, including ResNet, VGGNet, and GoogleNet. Data preprocessing involved rescaling (to 1/4, 1/2, and full original size). To enhance the robustness of the prediction model, data augmentation techniques were applied, including horizontal flipping, sharpness adjustment, brightness modification, and random rotation.\u003c/p\u003e \u003cp\u003eTo evaluate the effectiveness of sound information in the AI model, we incorporated joint noise information into the feature vector. We recorded objective noise (clinician-detected crepitus) and subjective noise reported by patients for each image. While crepitus data was collected for the entire dataset, subjective noise data was collected for 48.13% of the samples. Both crepitus and three types of patient-reported subjective sounds were binarized and encoded as four-dimensional vectors, with 1 indicating the presence of a sound and 0 indicating its absence. The integration of feature vectors and sound information was achieved by merging binarized sound information with the feature vectors derived from the TMJ panoramic images through the pre-trained models. A single-layer neural network used the merged vectors to perform the final diagnosis. To improve the performance of the AI model, experiments were conducted on six models by varying their inputs. The six models were:\u003c/p\u003e \u003cp\u003e \u003cul\u003e \u003cli\u003e \u003cp\u003eModel 1: image only\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eModel 2: image and crepitus\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eModel 3: image and subjective sounds (P, R, PR, I)\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eModel 4: image, crepitus and subjective sounds (P, R, PR, I)\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eModel 5: image and certain subjective sounds (R, PR)\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eModel 6: image, crepitus and certain subjective sounds (R, PR)\u003c/p\u003e \u003c/li\u003e \u003c/ul\u003e \u003c/p\u003e \u003cp\u003eThe structure of integrated AI model architecture was shown in Fig.\u0026nbsp;4\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec7\" class=\"Section2\"\u003e \u003ch2\u003eEvaluation of AI models' performance and clinical usability\u003c/h2\u003e \u003cp\u003eThe diagnostic performance of the AI model was evaluated by splitting the dataset into 60% training data and 40% test data. We conducted three repetitions of experiments with different training/test samplings. Due to class imbalance in the dataset, we used the F1-score instead of accuracy for performance evaluation. To train the AI model, we used the stochastic gradient descent (SGD) optimizer with a learning rate of 1.0 \u0026times; 10\u003csup\u003e\u0026ndash;3\u003c/sup\u003e. We trained AI models for a total of 300 epochs. The design, training, and evaluation of the diagnostic model were implemented using Python, with the PyTorch library utilized for designing and training the deep learning models. The saliency map was used to interpret and visualize the regions of the image considered by the AI while making predictions.\u003c/p\u003e \u003cp\u003eFor the clinical usability evaluation, 100 images labeled as \"D\" and 100 images labeled as \"N\" were randomly selected and mixed. These images were then reviewed by eight orofacial pain specialists. Each specialist evaluated the images twice, with and without noise joint information, at one-week intervals.\u003c/p\u003e \u003c/div\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eAuthor contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eE.C. contributed to the conception, design, data acquisition, analysis, and interpretation, drafted and critically revised the manuscript. Y.S. contributed to the design, data analysis, and interpretation, and critically revised the manuscript. S.T.K. contributed to the conception, design, critically revised the manuscript. S.S., K.L. and T.A. contributed to data analysis, interpretation, and original draft. Y.S. contributed to the design, data analysis, and interpretation, and critically revised the manuscript. R.L. and S.K. contributed to data acquisition. All authors gave final approval and agreed to be accountable for all aspects of the work.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData availability\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe data that support the findings of this study are available from the corresponding author upon reasonable request.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis study was supported by the Yonsei University College of Dentistry Fund (6-2024-0005) and the National Research Foundation of Korea (NRF) grant funded by the Ministry of Science and ICT (MSIT) of Korea (RS-2023-00208412).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting interests\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors declare no competing interests.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eWang, X., Zhang, J., Gan, Y. \u0026amp; Zhou, Y. Current understanding of pathogenesis and treatment of TMJ osteoarthritis. \u003cem\u003eJ. Dent. Res.\u003c/em\u003e \u003cb\u003e94\u003c/b\u003e, 666\u0026ndash;673 (2015).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSong, H., Lee, J. Y., Huh, K. H. \u0026amp; Park, J. W. Long-term changes of temporomandibular joint osteoarthritis on computed tomography. \u003cem\u003eSci. Rep.\u003c/em\u003e \u003cb\u003e10\u003c/b\u003e, 6731. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/s41598-020-63493-8\u003c/span\u003e\u003cspan address=\"10.1038/s41598-020-63493-8\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKim, K., Wojczyńska, A. \u0026amp; Lee, J. Y. The incidence of osteoarthritic change on computed tomography of Korean temporomandibular disorder patients diagnosed by RDC/TMD; a retrospective study. \u003cem\u003eActa Odontol. Scand.\u003c/em\u003e \u003cb\u003e74\u003c/b\u003e, 337\u0026ndash;342. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3109/00016357.2015.1136678\u003c/span\u003e\u003cspan address=\"10.3109/00016357.2015.1136678\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2016).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSchiffman, E. L. et al. The research diagnostic criteria for temporomandibular disorders. I: overview and methodology for assessment of validity. \u003cem\u003eJ. Orofac. Pain\u003c/em\u003e. \u003cb\u003e24\u003c/b\u003e, 7 (2010).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAlmășan, O., Leucuța, D. C., Hedeșiu, M., Mureșanu, S. \u0026amp; Popa, Ș. L. Temporomandibular joint osteoarthritis diagnosis employing artificial intelligence: systematic review and meta-analysis. \u003cem\u003eJ. Clin. Med.\u003c/em\u003e \u003cb\u003e12\u003c/b\u003e, 942. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/jcm12030942\u003c/span\u003e\u003cspan address=\"10.3390/jcm12030942\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eManfredini, D., Basso, D., Salmaso, L. \u0026amp; Guarda-Nardini, L. Temporomandibular joint click sound and magnetic resonance-depicted disk position: which relationship? \u003cem\u003eJ. Dent.\u003c/em\u003e \u003cb\u003e36\u003c/b\u003e, 256\u0026ndash;260. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.jdent.2008.01.002\u003c/span\u003e\u003cspan address=\"10.1016/j.jdent.2008.01.002\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2008).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWiese, M. et al. Association between temporomandibular joint symptoms, signs, and clinical diagnosis using the RDC/TMD and radiographic findings in temporomandibular joint tomograms. \u003cem\u003eJ. Orofac. Pain\u003c/em\u003e. \u003cb\u003e22\u003c/b\u003e, 239\u0026ndash;251 (2008).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSchiffman, E. et al. Diagnostic criteria for temporomandibular disorders (DC/TMD) for clinical and research applications: recommendations of the international RDC/TMD consortium network and orofacial pain special interest group. \u003cem\u003eJ. Oral Facial Pain Headache\u003c/em\u003e. \u003cb\u003e28\u003c/b\u003e, 6\u0026ndash;27. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.11607/jop.1151\u003c/span\u003e\u003cspan address=\"10.11607/jop.1151\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2014).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAbrahamsson, A. K. et al. Frequency of temporomandibular joint osteoarthritis and related symptoms in a hand osteoarthritis cohort. \u003cem\u003eOsteoarthr. Cartil.\u003c/em\u003e \u003cb\u003e25\u003c/b\u003e, 654\u0026ndash;657 (2017).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eIm, Y. G. et al. Diagnostic accuracy and reliability of panoramic temporomandibular joint (TMJ) radiography to detect bony lesions in patients with TMJ osteoarthritis. \u003cem\u003eJ. Dent. Sci.\u003c/em\u003e \u003cb\u003e13\u003c/b\u003e, 396\u0026ndash;404 (2018).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHoney, O. B. et al. Accuracy of cone-beam computed tomography imaging of the temporomandibular joint: comparisons with panoramic radiology and linear tomography. \u003cem\u003eAm. J. Orthod. Dentofac. Orthop.\u003c/em\u003e \u003cb\u003e132\u003c/b\u003e, 429\u0026ndash;438. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.ajodo.2005.10.032\u003c/span\u003e\u003cspan address=\"10.1016/j.ajodo.2005.10.032\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2007).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAhmad, M. et al. Research diagnostic criteria for temporomandibular disorders (RDC/TMD): development of image analysis criteria and examiner reliability for image analysis. \u003cem\u003eOral Surg. Oral Med. Oral Pathol. Oral Radiol. Endod\u003c/em\u003e. \u003cb\u003e107\u003c/b\u003e, 844\u0026ndash;860 (2009).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChoi, E., Kim, D., Lee, J. Y. \u0026amp; Park, H. K. Artificial intelligence in detecting temporomandibular joint osteoarthritis on orthopantomogram. \u003cem\u003eSci. Rep.\u003c/em\u003e \u003cb\u003e11\u003c/b\u003e, 10246. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/s41598-021-89742-y\u003c/span\u003e\u003cspan address=\"10.1038/s41598-021-89742-y\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2021).\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"Temporomandibular joint osteoarthritis, Artificial intelligence, Temporomandibular joint panoramic radiography, Joint noise","lastPublishedDoi":"10.21203/rs.3.rs-5086242/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-5086242/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eThis study aimed to develop an artificial intelligence (AI) model for the screening of temporomandibular joint osteoarthritis (TMJ OA) using temporomandibular joint (TMJ) panoramic radiography and joint noise data. A total of 2,631 TMJ panoramic images were collected, resulting in a final dataset of 3,908 images (2,127 normal (N) and 1,781 TMJ OA (D)) after excluding indeterminate cases and errors. AI models using GoogleNet were evaluated with six different combinations of image data, clinician-detected crepitus, and patient-reported joint noise. The model that integrated all joint noise data with imaging, demonstrated the highest performance, achieving an F1-score of 0.72. Another model, which incorporated both imaging and crepitus, also achieved the same F1-score but had lower D recall (0.55 vs. 0.67) and N precision (0.71 vs. 0.74). The AI models outperformed orofacial pain specialists when provided with imaging alone or in combination with all joint noise data. These findings suggest that AI-enhanced TMJ OA diagnosis using TMJ panoramic radiography and joint noise data offers a promising approach for early detection and improved patient care. The results underscore AI's capability to integrate diverse diagnostic factors, providing a comprehensive and accurate assessment that surpasses traditional methods.\u003c/p\u003e","manuscriptTitle":"Artificial intelligence-enhanced diagnosis of temporomandibular joint osteoarthritis using temporomandibular joint panoramic radiography and joint noise data","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-11-27 04:51:22","doi":"10.21203/rs.3.rs-5086242/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2024-11-07T06:42:28+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2024-10-27T17:16:38+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"21197951010317772852043246042934220344","date":"2024-10-25T05:31:21+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"68702283220824262291589189953935904539","date":"2024-10-24T16:44:15+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2024-10-11T17:50:00+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"166015663754743450037479272881457532887","date":"2024-09-29T23:01:46+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2024-09-28T13:09:18+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2024-09-28T06:34:55+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2024-09-25T06:07:15+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2024-09-25T06:03:30+00:00","index":"","fulltext":""},{"type":"submitted","content":"Scientific Reports","date":"2024-09-14T00:22:47+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"8d1b253c-15c6-4355-85e6-6fe4095fed29","owner":[],"postedDate":"November 27th, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[{"id":39932928,"name":"Health sciences/Diseases"},{"id":39932929,"name":"Health sciences/Medical research"}],"tags":[],"updatedAt":"2025-01-20T16:05:58+00:00","versionOfRecord":{"articleIdentity":"rs-5086242","link":"https://doi.org/10.1038/s41598-024-83750-4","journal":{"identity":"scientific-reports","isVorOnly":false,"title":"Scientific Reports"},"publishedOn":"2025-01-13 15:57:22","publishedOnDateReadable":"January 13th, 2025"},"versionCreatedAt":"2024-11-27 04:51:22","video":"","vorDoi":"10.1038/s41598-024-83750-4","vorDoiUrl":"https://doi.org/10.1038/s41598-024-83750-4","workflowStages":[]},"version":"v1","identity":"rs-5086242","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-5086242","identity":"rs-5086242","version":["v1"]},"buildId":"qtupq5eGEP_6zYnWcrvyt","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}