https://orcid.org/0000-0002-7030-569X
Figures
Abstract
Background
Artificial intelligence (AI) algorithms have been applied to diagnose temporomandibular disorders (TMDs). However, studies have used different patient selection criteria, disease subtypes, input data, and outcome measures. Resultantly, the performance of the AI models varies.
Objective
This study aimed to systematically summarize the current literature on the application of AI technologies for diagnosis of different TMD subtypes, evaluate the quality of these studies, and assess the diagnostic accuracy of existing AI models.
Materials and methods
The study protocol was carried out based on the preferred reporting items for systematic review and meta-analysis protocols (PRISMA). The PubMed, Embase, and Web of Science databases were searched to find relevant articles from database inception to June 2022. Studies that used AI algorithms to diagnose at least one subtype of TMD and those that assessed the performance of AI algorithms were included. We excluded studies on orofacial pain that were not directly related to the TMD, such as studies on atypical facial pain and neuropathic pain, editorials, book chapters, and excerpts without detailed empirical data. The risk of bias was assessed using the QUADAS-2 tool. We used Grading of Recommendations, Assessment, Development, and Evaluations (GRADE) to provide certainty of evidence.
Results
A total of 17 articles for automated diagnosis of masticatory muscle disorders, TMJ osteoarthrosis, internal derangement, and disc perforation were included; they were retrospective studies, case-control studies, cohort studies, and a pilot study. Seven studies were subjected to a meta-analysis for diagnostic accuracy. According to the GRADE, the certainty of evidence was very low. The performance of the AI models had accuracy and specificity ranging from 84% to 99.9% and 73% to 100%, respectively. The pooled accuracy was 0.91 (95% CI 0.76–0.99), I2 = 97% (95% CI 0.96–0.98), p < 0.001.
Conclusions
Various AI algorithms developed for diagnosing TMDs may provide additional clinical expertise to increase diagnostic accuracy. However, it should be noted that a high risk of bias was present in the included studies. Also, certainty of evidence was very low. Future research of higher quality is strongly recommended.
Citation: Jha N, Lee K-s, Kim Y-J (2022) Diagnosis of temporomandibular disorders using artificial intelligence technologies: A systematic review and meta-analysis. PLoS ONE 17(8): e0272715. https://doi.org/10.1371/journal.pone.0272715
Editor: Essam Al-Moraissi, Thamar University, Faculty of Dentistry, YEMEN
Received: January 20, 2022; Accepted: July 25, 2022; Published: August 18, 2022
Copyright: © 2022 Jha et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Data Availability: All relevant data are within the paper and its Supporting Information files.
Funding: YES-The National Research Foundation of Korea funded by the Ministry of Science and ICT of South Korea (grant 2019R1C1C1009881 and 2022R1H1A2011172) supported in the collection, analysis, and interpretation of data and in writing the manuscript.
Competing interests: The authors have declared that no competing interests exist.
Introduction
Temporomandibular disorders (TMDs) can cause pain and dysfunction in the temporomandibular joints (TMJs) and masticatory muscles. TMDs are the second most common musculoskeletal conditions and include various symptoms, such as decreased range of motion, joint sound, and mouth opening deviation [1]. TMDs can be classified as pain-related disorders, which include myalgia and arthralgia, and intra-articular disorders, which include internal derangement and degenerative joint disease (DJD) [2].
The etiology of TMDs is considered multifactorial, with biological, behavioral, and psychosocial factors contributing independently or as interrelated factors [3, 4]. Moreover, comorbidities, such as cardiovascular diseases, osteoarthritis, tinnitus, sinusitis, and thyroid disorders, are associated with disease onset and progression [5–7]. Therefore, diagnosis of TMDs requires a comprehensive evaluation of the patients’ signs and symptoms (acquired through clinical examination and medical image analysis) and behavioral and psychosocial factors [2, 8]. Subsequently, the complex nature of TMDs makes diagnosis difficult.
Currently, the most widely accepted diagnostic criteria is the Diagnostic Criteria for Temporomandibular Disorders (DC-TMD) [2] which was developed on the basis of large-scale international studies and data analyses since the 1990s. The DC-TMD comprises two axes, Axis I and Axis II, which include diagnostic standards for differentiating pain-related TMDs and intra-articular disorders (Axis I) and assessing jaw function and behavioral and psychosocial factors (Axis II).
Despite the popularity of the DC-TMD, it has limitations in terms of its diagnostic accuracy. Several subtypes of internal derangement, such as disc displacement with reduction, with reduction and locking, and without reduction, showed low sensitivity (0.34–0.54). Similarly, low sensitivity (0.55) and specificity (0.61) were observed for DJD. Further, the interexaminer reliability is relatively low for internal derangement and DJD [2]. Screening tools, such as surveys to determine patients’ symptoms, are expensive and time-consuming and place a burden on clinicians.
Advancements in artificial intelligence (AI) technologies have led to major developments in the healthcare industry. The Merriam–Webster dictionary defines AI as ‘the capability of a machine to imitate intelligent human behavior.’ It essentially refers to the simulation of human intelligence processes using computer systems. Generally, AI systems are trained using large amounts of input data. Patterns are learned from these data and then used to predict the outcome of new instances. AI algorithms are increasingly applied in patient diagnoses, especially for detecting and classifying lesions, such as skin cancers [9], diabetic retinopathy [10], brain tumors [11], and dental caries [12], using medical diagnostic images [13]. Additionally, other data types, such as electronic medical records in the form of text [14], voice [15], and sound [16] are used to develop diagnostic tools to support clinicians in decision-making.
Recently, various AI algorithms have been applied to image and nonimage data for TMDs diagnosis [17–21]. However, studies on the use of AI for TMD diagnosis have used different patient selection criteria, disease subtypes, input data used for diagnosis, and outcome measures for performance evaluation. Moreover, the accuracy of the AI models varies. To the best of our knowledge, there has been no systematic review till date that summarizes such findings. Therefore, this study aimed to systematically summarize the current literature on the application of AI technologies for diagnosis of different TMD subtypes—both muscular and articular conditions—evaluate the quality of these studies and assess the diagnostic accuracy of existing AI models.
Materials and methods
This systematic review and meta-analysis was conducted and reported in accordance with the Preferred Reporting Items for Systematic Review and Meta-analysis (PRISMA) 2020 guidelines (S1 and S2 Tables) [22].
Research questions
This systematic review and meta-analysis was conducted to answer the following question: “How accurate are the AI algorithms for the diagnosis of TMDs?” The focused question was further classified as follows:
- Which data were used for developing algorithms for TMD diagnosis?
- Which AI techniques were used for TMD diagnosis?
- Which features were used for TMD diagnosis?
- Which outcome measures were used for assessing the model performance?
Further, the research question was formatted using the Population, Intervention, Comparison, and Outcome framework (Table 1).
Information sources and search strategy
Our search algorithm comprised the PubMed, EMBASE, and Web of Science databases. A combination of the following terms was used: “artificial intelligence” OR “neural network” OR “machine learning” OR “deep learning” OR/AND “TMJ osteoarthritis” OR “temporomandibular joint osteoarthritis” OR “temporomandibular disorders” OR “masticatory muscle disorders” OR “TMDs” OR “TMJ disorder” OR “temporomandibular joint disorders” OR “TMJ arthritis” OR “temporomandibular joint arthritis” OR “progressive condylar resorption” OR “degenerative joint disease” OR “temporomandibular joint disease” OR “TMJ disease” OR “idiopathic condylar resorption” OR “juvenile idiopathic arthritis.” No start date was used, whereas the end date was June 30, 2022. Table 2 includes the search strategy for each database.
Eligibility criteria, study selection, and data collection
We included original studies published in scientific journals whose full texts were available. The inclusion criteria were as follows: (a) use of AI algorithms to diagnose at least one subtype of TMDs; (b) the performance of the developed AI algorithms was assessed; (c) no limit on the participants in terms of gender, age, or ethnicity; and (d) were written in English. The exclusion criteria were as follows: (a) studies on orofacial pain that is not directly related to the TMJ, such as atypical facial pain and neuropathic pain; (b) studies on TMJ that were unrelated to disease diagnosis; (c) editorials, comments, book chapters, and excerpts without detailed empirical data; and (d) studies not written in English.
To determine the final eligibility, the two investigators (YJK and NJ) independently assessed the full text of studies. Conflicts between the reviewers was resolved by the involvement of a third investigator (KSL). Then, two investigators, NJ and YJK, independently extracted and formulated the data, such as input data used for TMD diagnosis, AI algorithms used, and performance measures. Any discrepancies were resolved through discussion.
Risk of bias assessment
The selected articles were critically assessed and scored independently by two investigators (YJK and NJ). Quality assessment of the studies was based on the Quality Assessment of Diagnostic Accuracy Studies (QUADAS-2) [23]. The QUADAS tool was first developed in 2003 for systematic reviews of diagnostic accuracy studies and later updated to QUADAS-2. It comprises four components: patient selection, index test, reference standard, and flow and timing. Each component is assessed for the risk of bias. The first three components are also assessed for concerns about the applicability of each component [23]. The quality was rated as high, low, or unclear. Conflicts between the reviewers was resolved by the involvement of a third investigator (KSL).
Certainty of evidence assessment
We used Grading of Recommendations, Assessment, Development, and Evaluation (GRADE) [24] to evaluate the quality of evidence of studies for which meta-analysis was performed. Each outcome gets a rating on the quality of evidence of high, moderate, low, or very low within five domains- risk of bias, imprecision, inconsistency, indirectness, and publication bias.
Statistical analysis
Meta-analysis of diagnostic accuracy was conducted using the Hartung–Knapp–Sidik–Jonkman method for random-effects models. The accuracy estimates were transformed using the Freeman–Tukey double arcsine method. Heterogeneity was quantified using the I2 statistic, which is the percentage of total variation across studies due to heterogeneity rather than chance. All analyses were conducted using R v.4.0.4 (R Project for Statistical Computing) with the Meta package.
Results
Study selection
The initial database search yielded 1923 studies. After removing duplicate studies, 985 articles were screened for inclusion, of which 32 studies corresponded to TMD diagnosis using AI. However, 15 of these 32 articles were excluded due to various reasons, such as book chapters, studies with a focus on creating a web system repository for neural data storage, studies related to TMJ movement and anatomy, excluding diagnosis, studies related to facial pain syndrome as a differential diagnosis, and studies related to robotics (S3 Table). Finally, 17 articles met our eligibility criteria and were included in this systematic review (Fig 1).
Risk of bias assessment of the included studies
Fig 2 summarizes the study biases as high, low, or unclear. The patient selection bias potential was low in 11 out of 17 studies [17, 19, 21, 25–32] and high in 6 out of 17 studies [18, 20, 33–36]. A high risk of bias in patient selection was present due to the inclusion of case-control studies. However, the applicability concerns for patient selection were assessed as low for these studies because selection bias was overcome using case-control matching. Regarding the reference test and flow and timing domains, 17 out of 17 studies were considered to have a low degree of bias and low degree of applicability (S1 Fig). Index test was reported unclear for 13 out of 17 studies due to a lack of information on threshold values.
Certainty of evidence assessment of the included studies
Of the 7 studies considered for meta-analysis, 2 studies had invalid outcomes for the test of diagnostic accuracy. Therefore, 5 studies were included for the GRADE analysis [17, 18, 28, 33, 34]. According to the GRADE, the Risk of bias was considered serious as it was high for three studies [18, 33, 34]. The factor of imprecision was considered very serious because the number of subjects was less than 1000 [17, 18, 33, 34]. Therefore, the certainty of evidence was concluded as very low (S4 Table).
Characteristics of the included studies
The types of studies included were mostly retrospective studies [17, 20, 21, 25, 26, 28, 30–33], 3 case-control studies [18, 19, 29, 34], 2 case-control cohort studies [35, 36], and 1 pilot study [27]. Table 3 shows the characteristics of the included studies.
Sex distribution indicated higher number of female subjects than male subjects for most of the studies [17, 21, 27, 30, 31, 35, 36]. Image and nonimage data were used, and medical diagnostic imaging modalities, such as CBCT [17, 18, 20, 26, 33], high-resolution CBCT (HR-CBCT) [34–36], MRI [21, 33], and panoramic radiography [28] were used. Other types of image data included infrared cameras with a motion-tracking system [27] and high-resolution video cameras [19]. Nonimage data included medical records, such as patients’ symptoms [25, 29, 31, 32]. The most frequently used method was convolutional neural networks (CNNs; 7 studies), followed by artificial neural networks (ANN; 5 studies), and decision trees (4 studies). Other techniques included Bayesian networks (3 studies), support vector machines (SVMs; 3 studies), K-nearest neighbors (KNNs; 2 studies), and natural language processing (NLP; 1 study). Some studies used several machine-learning algorithms and compared the results.
Meta-analysis
The diagnostic accuracy was 0.69–1.00, and the pooled accuracy was 0.91 (95% CI 0.76–0.99), I2 = 97% (95% CI 0.96–0.98), p < 0.001 (Fig 3). The study with the lowest accuracy had multiple classes of condylar shape in patients with DJD in which the classes represented varying degrees of condylar resorption and remodeling [33].
Discussion
Diagnosis of TMDs can be complex as patients present with various symptoms according to subtypes, thus requiring clinical expertise. Various studies have diagnosed TMDs using AI to facilitate diagnosis and support clinical decisions. However, the accuracy of the developed models varied greatly depending on the type of data used, dataset size, and algorithms used for developing the model.
Among the subtypes of the TMDs, TMJOA was found to be the most studied type of TMD in this systematic literature review. One of the possible reasons is that TMJOA is an advanced form of disease that occurs after disc displacement, and it has a significant effect on occlusion and facial appearance. Deep-learning algorithms were used to diagnose TMJOA by detecting the changes in the condyle shape using CBCT images [18, 20, 33]. Lee et al. developed an automated diagnostic tool for detecting TMJOA based on the Diagnostic Criteria for TMDs [17]. Kim et al. used panoramic radiographs to automatically detect the condyles and classify osteoarthritis [28]. Although panoramic radiographs are not considered the standard imaging technique in the diagnosis of TMJOA [4], the AI model showed accuracy, sensitivity, and specificity of 0.84, 0.54, and 0.94, respectively, for diagnosing bony abnormality [28]. Machine-learning methods were used to examine correlations between the biomarkers, and condylar shape changes were investigated to increase diagnostic sensitivity [34–36]. Radiomics features were extracted from high-resolution CBCT scans to detect early bony changes [35, 36].
All studies on TMJOA used image data to analyze mandibular condyle shapes [17–20, 28–31, 33–36], and CBCT was the most commonly used imaging modality. Accurate assessment of bony changes is possible using CBCT; thus, it is considered the gold standard for TMJOA [37]. HR-CBCT scans at a submillimeter resolution with voxel size as low as 80 μm [38]. Compared with micro-CT, it allows observing subtle changes in the trabecular pattern of the condyle [35, 39]. The accuracy of the AI models used in these studies ranged from 80% to 90%, indicating their high reliability. These results are similar to the conventional studies involving human experts to diagnose TMJOA using CBCT [40, 41]. MRI was the most frequently used imaging method for the diagnosis of internal derangements and disc perforations [21, 30]. Other data include jaw movement records [27]. Bas et al. used clinical symptoms and diagnoses to predict the subtypes of internal derangements using ANNs [25]. We provide a brief explanation of techniques used in each study below.
ANN is a popular AI model that includes one input layer, two or three hidden layers, and one output layer. ANN training begins by randomly assigning weights as small numbers near 0 and iterating the feedforward and backpropagation algorithms until certain criteria are met to accurately predict the final output [42].
Deep learning is a subgroup of ANNs that involves many hidden layers. CNNs are a type of deep learning algorithms that have been developed for image data analysis. CNNs can be used for medical image analysis by performing tasks such as classification, which identifies input image data as pretrained classes (such as disease or normal), detection, which locates the region of interest (i.e. abnormal area), and segmentation, which identifies regions of interest as pixel-wise boundaries [43–45].
Decision trees are popular tools that present results in a tree structure that can be easily interpreted, are less time-consuming, and can help understand the interactions among different features [46]. Decision tree algorithms were used by four studies in various forms, such as random forest [26, 27, 30], light gradient boosting machine, and XGBoost [35].
Bayesian networks are a group of techniques connecting statistics and machine learning applicable to complex systems, which can leverage smaller data sizes compared with other machine-learning algorithms [47]. Further, large probability distributions can be compactly represented using Bayesian networks [48]. They comprise factorizing a probability distribution and a corresponding directed acrylic graph (DAG). The DAG presents a cause–effect relationship among nodes [21, 48]. Bayesian networks have many forms, including naïve Bayes (supervised classification) [45], greedy search-and-score [21], and Bayesian belief network path condition [21].
SVMs have been recently developed and are useful techniques in pattern recognition and classification studies [49]. Algorithm consideration, i.e., selecting a kernel/learning function, made in advance, can improve the performance of SVMs. This technique involves the nonlinear mapping of input vectors in a high-dimensional feature space to construct a linear decision surface [49].
KNN is one of the simplest classification methods wherein the samples are divided into training and testing groups. Training is performed with known labels, following which test samples are predicted using the learned model. The training and testing data need not be identical for KNN [50].
NLP is a subfield of AI that is used to decode human language into computer language [31]. Hospital data in the form of clinical history, radiology reports, and physical examination findings are available from clinical databases; these can be interpreted with computational linguistics using AI-assisted NLP systems. Free text can be organized into structured data [31, 51], which reduces labor-intensive and error-prone administrative demands.
Feature extraction techniques such as gray-level co-occurrence matrix, gray-level run-length matrix [36], local binary patterns [26], and histograms of oriented gradients [26] are used as image-processing techniques to automatically analyze texture, shape, and color changes within images. Feature selection is an important step in classification [52]. Different feature extraction algorithms can be sequentially applied to extract feature matrices for individual images. Following this method, feature matrix classification is performed using algorithms, such as SVM and KNN [52]. Principal component analysis (PCA) is a mathematical algorithm used to identify variations in data that simultaneously reduces their dimensionality, creating sample plotting, and identifying similarities and differences within a group of simple tasks [53].
Regarding the risk of bias assessment, this study used the QUADAS-2 tool recommended for systematic reviews of diagnostic accuracy by the Agency for Healthcare Research and Quality, Cochrane Collaboration [54]. We could have used the Cochrane tool for Risk Of Bias due to Missing Evidence in a synthesis. However, this tool was intended for risk of bias assessment for the meta-analyses of the effects of interventions [55]. Some of the included studies showed a high risk of bias in the patient selection domain because they were case-control studies. Other domains showed a low risk of bias and low risk of applicability concerns for all included studies.
Regardless of the possible risk of patient selection bias, most of the included studies reported high performance of the AI models showing a pooled accuracy of 0.91. However, there was a concern about the quality of evidence due to the small number of subjects included in the studies. Moreover, apart from the quality of the evidence, most studies lacked robust validation mechanisms. Validation, i.e., model performance evaluation, may be evaluated using data used for model development (internal) or from separate data that is not used for model development (external) [56]. Crossvalidation or validating from similar data sources may introduce accuracy bias [57]. External validation mechanisms, such as cohort studies, data collection from various institutions, prospective data [58], and data from different sites [56], are needed to improve the accuracy, quality, and generalizability of AI models.
Accuracy of traditional diagnostic tools for TMDs varies greatly. A systematic review on the diagnostic accuracy of clinical diagnostic tests and signs of TMD reported sensitivity and specificity of 2–89% and 14–97%, respectively [59]. The diagnostic accuracy varied according to the disease subtype and diagnostic test and signs used. In contrast, medical imaging modalities such as CT and MRI, which are regarded as gold standards for diagnosis of osteoarthritis and internal derangement, respectively, have shown a high examiner reliability [60]. Latest AI technologies have been introduced to support clinicians in diagnosing TMDs using various types of data, such as medical diagnostic images, video images, radiomics features, jaw movement tracking, electronic medical records (EMR), and biomarkers. These may contribute to the increased diagnostic accuracy.
This study has a few limitations. Most of the included studies have reported the model performance in terms of sensitivity, specificity, accuracy, recall, and R1. However, they did not provide raw data for meta-analysis of sensitivity and specificity, except for one study [14]. Therefore, only accuracy could be calculated in the meta-analysis. Additionally, the accuracies of the included studies showed high heterogeneity because the AI algorithms were developed for different TMD subtypes, thus the number of classes in the output and the criteria for accurate prediction varied among studies. Another limitation is that the study protocol was not registered in PROSPERO, and the transparency of this study could be affected. Lastly, we omitted abstracts and conference proceedings in our review and only used English articles selected from major databases, which collectively may exclude relevant studies published in other languages.
Conclusions
The results of this study suggest that AI algorithms developed for automated TMD diagnosis can be used as a decision support tool for clinicians. In addition to the medical diagnostic images, various input data types, such as EMR, biomarkers, and radiomics features may help increase the diagnostic accuracy of TMDs. However, a high risk of bias in patient selection was present due to the inclusion of case-control studies. Most of the studies used a small training dataset and lacked external validation. Additionally, a significant heterogeneity was observed among the studies included for meta-analysis of diagnostic accuracy. The certainty of evidence was concluded as very low. Further studies with a larger dataset to prevent overfitting and ensure generalizability of developed models are warranted.
Supporting information
S1 Fig. Quality assessment (QUADAS-2) summary table for individual studies.
https://doi.org/10.1371/journal.pone.0272715.s001
(TIF)
S1 Table. PRISMA 2020 for abstracts checklist.
https://doi.org/10.1371/journal.pone.0272715.s002
(DOCX)
S4 Table. GRADE assessment of the level of evidence for all included studies.
https://doi.org/10.1371/journal.pone.0272715.s005
(DOCX)
References
- 1.
National Institute of Dental and Craniofacial Research. 2018. Facial Pain and its signs and symptoms [accessed 2021 June 10]. http://nidcr.nih.gov
- 2. Schiffman E, Ohrbach R, Truelove E, Look J, Anderson G, Goulet JP, et al. International RDC/TMDs Consortium Network, International association for Dental Research; Orofacial Pain Special Interest Group, International Association for the Study of Pain. Diagnostic Criteria for Temporomandibular Disorders (DC/TMDs) for Clinical and Research Applications: recommendations of the International RDC/TMDs Consortium Network and Orofacial Pain Special Interest Group. J Oral Facial Pain Headache. 2014;28: 6–27.
- 3.
de leeuw R, Klasser GD. Orofacial Pain-Guidelines for Assessment, Diagnosis and Management. The American Academy of Orofacial Pain, 2018; 5th ed.
- 4. Schiffman E, Truelove E, Ohrbach R, Anderson G, John MT, List T, et al. The research diagnostic criteria for temporomandibular disorders. I: overview and methodology for assessment of validity. J Orofac Pain. 2010;24(1): 7–24. pmid:20213028
- 5. Burris JL, Evans DR, Carlson CR. Psychological correlates of medical comorbidities in patients with temporomandibular disorders. J Am Dent Assoc. 2010;141: 22–31. pmid:20045818
- 6. Skog C, Fjellner J, Ekberg E, Häggman-Henrikson B. Tinnitus as a comorbidity to temporomandibular disorders-A systematic review. J Oral Rehabil. 2019;46: 87–99. pmid:30126027
- 7. Song HS, Shin JS, Lee J, Lee YJ, Kim MR, Cho JH, et al. Association between temporomandibular disorders, chronic diseases, and ophthalmologic and otolaryngologic disorders in Korean adults: A cross-sectional study. PLoS One. 2018; 13(1): e0191336. pmid:29385182
- 8. McKinney MW, Lundeen TF, Turner SP, Levitt SR. Chronic TM disorder and non-TM disorder pain: a comparison of behavioral and psychological characteristics. Cranio. 1990;8(1): 40–46. pmid:2098186
- 9. Esteva A, Kuprel B, Novoa RA, Ko J, Swetter SM, Blau HM, et al. Dermatologist-level classification of skin cancer with deep neural networks. Nature. 2017;542(7639): 115–118. pmid:28117445
- 10. Wong TY, Bressler NM. Artificial intelligence with deep learning technology looks into diabetic retinopathy screening. Jama. 2016;316(22): 2366–2367. pmid:27898977
- 11.
Bhanumurthy MY, Anne K. An automated detection and segmentation of tumor in brain MRI using artificial intelligence. IEEE International Conference on Computational Intelligence and Computing Research; 2014; 1–6.
- 12. Lee JH, Kim DH, Jeong SN, Choi SH. Detection and diagnosis of dental caries using a deep learning-based convolutional neural network algorithm. J Dent. 2018;77: 106–111. pmid:30056118
- 13. Machoy ME, Szyszka-Sommerfeld L, Vegh A, Gedrange T, Woźniak K. The ways of using machine learning in dentistry. Adv Clin Exp Med. 2020; 29(3): 375–384. pmid:32207586
- 14. Liao KP, Cai T, Savova GK, Murphy SN, Karlson EW, Ananthakrishnan AN, et al. Development of phenotype algorithms using electronic medical records and incorporating natural language processing. BMJ. 2015;350: h1885. pmid:25911572
- 15.
Ouhmida A, Terrada O, Raihani A, Cherradi B, Hamida S. Voice-Based Deep Learning Medical Diagnosis System for Parkinson’s Disease Prediction. International Congress of Advanced Technology and Engineering (ICOTEN). 2021; 1–5.
- 16.
Singh SA, Majumder S, Mishra M. Classification of short unsegmented heart sound based on deep learning. 2019 IEEE International Instrumentation and Measurement Technology Conference (I2MTC). 2019; 1–6.
- 17. Lee KS, Kwak HJ, Oh JM, Jha N, Kim YJ, Kim W, et al. Automated Detection of TMJ Osteoarthritis Based on Artificial Intelligence. J Dent Res. 2020; 99(12): 1363–1367. pmid:32609562
- 18. Ribera NT, de Dumast P, Yatabe M, Ruellas A, Ioshida M, Paniagua B, et al. Shape variation analyzer: a classifier for temporomandibular joint damaged by osteoarthritis. Proc SPIE Int Soc Opt Eng. 2019; 10950: 1095021. pmid:31359900
- 19.
Ghodsi M, Sanei S, Hicks Y, Lee T, Dunne S. Detection of Temporomandibular Disorder from Facial Pattern. 15th International Conference on Digital Signal Processing, 2007; 151–154.
- 20. de Dumast P, Mirabel C, Paniagua B, Yatabe M, Ruellas A, Tubau N, et al. SVA: Shape variation analyzer. Proc SPIE Int Soc Opt Eng. 2018; 10578:105782H. pmid:29780198
- 21. Iwasaki H. Bayesian belief network analysis applied to determine the progression of temporomandibular disorders using MRI. Dentomaxillofac Radiol. 2015; 44(4):20140279. pmid:25472616
- 22. Page M J, McKenzie J E, Bossuyt P M, Boutron I, Hoffmann T C, Mulrow C D et al. The PRISMA 2020 statement: an updated guideline for reporting systematic reviews BMJ 2021; 372: n71. pmid:33782057
- 23. Whiting PF, Rutjes AW, Westwood ME, Mallett S, Deeks JJ, Reitsma JB, et al. QUADAS-2 Group. QUADAS-2: a revised tool for the quality assessment of diagnostic accuracy studies. Ann Intern Med. 2011 Oct 18;155(8): 529–536.
- 24. Guyatt GH, Oxman AD, Vist GE, Kunz R, Falck-Ytter Y, Alonso-Coello P, et al. GRADE Working Group. GRADE: an emerging consensus on rating quality of evidence and strength of recommendations. BMJ. 2008; 336(7650): 924–926.
- 25. Bas B, Ozgonenel O, Ozden B, Bekcioglu B, Bulut E, Kurt M. Use of artificial neural network in differentiation of subgroups of temporomandibular internal derangements: a preliminary study. J Oral Maxillofac Surg. 2012;70(1): 51–59. pmid:21802818
- 26. Haghnegahdar AA, Kolahi S, Khojastepour L, Tajeripour F. Diagnosis of Temporomandibular Disorders Using Local Binary Patterns. J Biomed Phys Eng. 2018;8(1): 87–96.
- 27. Calil BC, da Cunha DV, Vieira MF, de Oliveira Andrade A, Furtado DA, Bellomo Junior DP, et al. Identification of arthropathy and myopathy of the temporomandibular syndrome by biomechanical facial features. Biomed Eng Online. 2020;19(1): 22. pmid:32295597
- 28. Kim D, Choi E, Jeong HG, Chang J, Youm S. Expert system for Mandibular Condyle Detection and Osteoarthritis Classification in Panoramic Imaging Using R-CNN and CNN. Appl Sci. 2020; 10(21): 7464.
- 29. Radke JC, Ketcham R, Glassman B, Kull R. Artificial neural network learns to differentiate normal TMJs and nonreducing displaced discs after training on incisor-point chewing movements. Cranio. 2003;21(4): 259–264.
- 30. Kim JY, Kim D, Jeon KJ, Kim H, Huh JK. Using deep learning to predict temporomandibular joint disc perforation based on magnetic resonance imaging. Sci Rep. 2021;11(1):6680. pmid:33758266
- 31. Nam Y, Kim HG, Kho HS. Differential diagnosis of jaw pain using informatics technology. J Oral Rehabil. 2018;45(8): 581–588. pmid:29782036
- 32. Kreiner M, Viloria J. A novel artificial neural network for the diagnosis of orofacial pain and temporomandibular disorders. J Oral Rehabil. 2022. pmid:35722743
- 33. de Dumast P, Mirabel C, Cevidanes L, Ruellas A, Yatabe M, Ioshida M, et al. A web-based system for neural network based classification in temporomandibular joint osteoarthritis. Comput Med Imaging Graph. 2018;67: 45–54. pmid:29753964
- 34. Shoukri B, Prieto JC, Ruellas A, Yatabe M, Sugai J, Styner M, et al. Minimally Invasive Approach for Diagnosing TMJ Osteoarthritis. J Dent Res. 2019; 98(10): 1103–1111. pmid:31340134
- 35. Bianchi J, de Oliveira Ruellas AC, Gonçalves JR, Paniagua B, Prieto JC, Styner M, et al. Osteoarthritis of the Temporomandibular Joint can be diagnosed earlier using biomarkers and machine learning. Sci Rep. 2020;10(1): 8012. pmid:32415284
- 36. Bianchi J, Gonçalves JR, de Oliveira Ruellas AC, Ashman LM, Vimort JB, Yatabe M, et al. Quantitative bone imaging biomarkers to diagnose temporomandibular joint osteoarthritis. Int J Oral Maxillofac Surg. 2021;50(2): 227–235. pmid:32605824
- 37. Talmaceanu D, Lenghel M, Bolog N Hedesiu M, Buduru S, Rotar H, Baciut M, et al. Imaging modalities for temporomandibular joint disorders: An update. Clijul Med. 2018;91(3): 280–287. pmid:30093805
- 38. Paniagua B, Ruellas AC, Benavides E, Marron S, Woldford L, Cevidanes L. Validation of CBCT for the computation of textural biomarkers. Proc SPIE Int Soc Opt Eng. 2015;9417: 94171B. pmid:26085710
- 39. Bianchi J, Gonçalves JR, Ruellas ACO, Vimort JB, Yatabe M, Paniagua B, et al. Software comparison to analyze bone radiomics from high resolution CBCT scans of mandibular condyles. Dentomaxillofac Radiol. 2019; 48(6): 20190049. pmid:31075043
- 40. Yadav S, Palo L, Mahdian M, Upadhyay M, Tadinada A. Diagnostic accuracy of 2 cone-beam computed tomography protocols for detecting arthritic changes in temporomandibular joints. Am J Orthod Dentofacial Orthop. 2015 Mar;147(3): 339–344. pmid:25726401
- 41. Larheim TA, Abrahamsson AK, Kristensen M, Arvidsson LZ. Temporomandibular joint diagnostics using CBCT. Dentomaxillofac Radiol. 2015;44(1): 20140235 pmid:25369205
- 42.
Han J, Kamber M, Pei J. Data mining: concepts and techniques. 3rd ed. Elsevier 2011:744.
- 43. Kim M, Yun J, Cho Y, Shin K, Jang R, Bae HJ, et al. Deep Learning in Medical Imaging. Neurospine. 2019;16(4): 657–668 pmid:31905454
- 44. Soffer S, Ben-Cohen A, Shimon O, Amitai MM, Greenspan H, Klang E. Convolutional Neural Networks for Radiologic Images: A Radiologist’s Guide. Radiol. 2019; 290(3): 590–606. pmid:30694159
- 45.
Liu W, Anguelov D, Erhan D, Szegedy C, Reed S, Fu C-Y, et al. SSD: Single Shot MultiBox Detector. Cham: Springer International Publishing; 2016; 21–37.
- 46. Stiglic G, Kocbek S, Pernek I, Kokol P. Comprehensive decision tree models in bioinformatics. PLoS One. 2012;7(3): e33812. pmid:22479449
- 47. Becker AK, Dörr M, Felix SB, Frost F, Grabe HJ, Lerch MM, et al. From heterogeneous healthcare data to disease-specific biomarker networks: A hierarchical Bayesian network approach. PLoS Comput Biol. 2021;17(2): e1008735. pmid:33577591
- 48. Koski TJ, Noble J. A review of Bayesian networks and structure learning. Mathematica Applicanda. 2012; 40(1).
- 49. Cortes C, Vapnik V. Support-vector networks. Mach Learn. 1995; 20(3): 273–297.
- 50. Bian Z, Vong CM, Wong PK, Wang S. Fuzzy KNN Method With Adaptive Nearest Neighbors. IEEE Trans Cybern. 2020.
- 51. Juhn Y, Liu H. Artificial intelligence approaches using natural language processing to advance EHR-based clinical research. J Allergy Clin Immunol. 2020;145(2): 463–469. pmid:31883846
- 52. Öztürk S, Akdemir B. Application of Feature Extraction and Classification Methods for Histopathological Image using GLCM, LBP, LBGLCM, GLRLM and SFTA. Procedia Computer Science. 2018;132: 40–46.
- 53. Ringnér M. What is principal component analysis? Nat Biotechnol. 2008;26(3): 303–304. pmid:18327243
- 54. Reitsma JB, Rutjes AW, Whiting P, Vlassov VV, Leeflang MM, Deeks JJ. Assessing methodological quality. Cochrane handbook for systematic reviews of diagnostic test accuracy version. 2009 Oct 27;1(0): 1–28.
- 55. Higgins JP, Altman DG, Gøtzsche PC, Jüni P, Moher D, Oxman AD, et al. Cochrane Bias Methods Group; Cochrane Statistical Methods Group. The Cochrane Collaboration’s tool for assessing risk of bias in randomised trials. BMJ. 2011 Oct 18;343: d5928.
- 56. Park SH, Han K. Methodologic Guide for Evaluating Clinical Performance and Effect of Artificial Intelligence Technology for Medical Diagnosis and Prediction. Radiol. 2018; 286(3): 800–809. pmid:29309734
- 57. Kim DW, Jang HY, Kim KW, Shin Y, Park SH. Design Characteristics of Studies Reporting the Performance of Artificial Intelligence Algorithms for Diagnostic Analysis of Medical Images: Results from Recently Published Papers. Korean J Radiol. 2019; 20(3): 405–410. pmid:30799571
- 58. Farrah K, Young K, Tunis MC, Zhao L. Risk of bias tools in systematic reviews of health interventions: an analysis of PROSPERO-registered protocols. Syst Rev. 2019 Nov 15;8(1):280 pmid:31730014
- 59. Reneker J, Paz J, Petrosino C, Cook C. Diagnostic accuracy of clinical tests and signs of temporomandibular joint disorders: a systematic review of the literature. J Orthop Sports Phys Ther. 2011 Jun;41(6):408–416. pmid:21335932
- 60. Ahmad M, Hollender L, Anderson Q, Kartha K, Ohrbach R, Truelove EL, 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. 2009 Jun;107(6):844–860. pmid:19464658