View the CAT printer-friendly / share this CAT
Title Artificial Intelligence May Aid in Diagnosis of TMJ Osteoarthritis, but the Evidence Is Not Clear
Clinical Question For patients with TMJ osteoarthritis, can analysis of radiographs such as CT, CBCT, Panoramic, or MRI via artificial intelligence (AI) result in more accurate or earlier diagnosis when compared to interpretation by conventional radiologists?
Clinical Bottom Line While it is possible that artificial intelligence could achieve a moderate to high degree of accuracy for diagnosis of TMJ osteoarthritis, especially with CBCT images, there is not enough high-quality evidence to support use of this diagnostic tool at this point.
Best Evidence (you may view more info by clicking on the PubMed ID link)
PubMed ID Author / Year Patient Group Study type
(level of evidence)
#1) 36769590Almășan/20237 studies/10,077 images (Meta-analysis: 3 studies/5520 images)Meta-Analysis
Key resultsThis meta-analysis focused on the sensitivity and specificity of the diagnostic ability of artificial intelligence (AI) tools to detect TMJ osteoarthritis on panoramic radiographs compared to radiologic experts who also employed clinical information and CBCT to form a definitive diagnosis. The overall sensitivity of the AI model for detecting osteoarthritis was 0.76 (95% CI 0.35–0.95), p = 0.208 and the overall specificity was 0.79 (95% CI 0.75–0.83), p = 0.208. The threshold of p < 0.05 was not met for either metric. There were also four studies that focused on AI interpretation of CBCT that were included in the systematic review but not in the meta-analysis. The study by Bianchi found the best of four machine learning algorithms to have an accuracy and F1 score of 0.823. Another study by De Dumast found “a 91% agreement between the clinician and the SVA classifier.” Diagnostic investigation by Lee KS resulted in a precision and F1 score of 0.89 when indeterminate diagnoses were excluded. The fourth study that used CBCT as the imaging modality only compared two machine learning algorithms against each other, so that study was not relevant to the clinical question evaluated by this CAT.
Evidence Search TMJ AND (computed tomography OR CT OR CBCT OR MRI OR Panoramic) AND Artificial Intelligence AND osteoarthritis
Comments on
The Evidence
The meta-analysis by Almășan reported moderate-to-high sensitivity and specificity for AI diagnosis of TMJ osteoarthritis (OA), but the sensitivity had a wide range for its 95% CI, indicating the actual sensitivity may be as low as 0.35, which would be clinically unacceptable. Also, due to the exclusion of the four CBCT studies included in the accompanying systematic review, this result can only apply to AI’s ability to diagnose TMJ OA using panoramic radiography, thus greatly limiting its scope. The heterogeneity of the studies’ sensitivity was also a problem, with I2 = 96.41% (p < 0.001). The evidence was further weakened by the fact that all the included studies discarded diagnoses of “indeterminate for TMJ OA,” which artificially raised the pooled sensitivity of the AI diagnostic tool. Although the systematic review did a thorough search of the literature and multiple reviewers were used with a tie breaking procedure for disagreements, the exclusion criteria described in the review were vague. It appears the three studies selected for the meta-analysis were included because they all used panoramic radiography and a ResNet learning model. The excluded CBCT studies showed promising numbers that were all above 0.8 for precision/accuracy/F1 scores, but details of these studies such as sample size, confidence intervals, and p-values were left out, so the quality of the evidence cannot be readily evaluated. Overall, the evidence presented is not strong enough to support the widespread use of AI to aid in diagnosis of TMJ OA from panoramic radiographs and the evidence supporting use of AI on CBCT images demonstrates promise, but further study with more consistent use of the same or very similar learning models and inclusion of the diagnosis “indeterminate for TMJ OA” would aid in establishing a stronger case for the use of AI models.
Applicability It would be helpful for radiologists and clinicians to have a tool that allows for more accurate and earlier diagnosis of TMJ OA in order to better intervene and lower the associated morbidity. Unfortunately, the studies presented supporting the use of AI in TMJ osteoarthritis diagnosis provide us with insufficient evidence to influence the standard of care. Given the level of distrust some clinicians have with technology, it is extremely important that any AI tool, especially if used without the aid of a trained radiologist, have a very high specificity and sensitivity as well as quality scientific evidence to support it. More studies are required on the topic with standardized or similar machine learning algorithms and 3D techniques such as CBCT or MRI as the sole imaging modalities. Without an increase in the quality of evidence, it would be prudent to refrain from use of AI in the diagnosis of TMJ OA.
Specialty/Discipline (Oral Medicine/Pathology/Radiology) (General Dentistry) (Orthodontics) (Prosthodontics)
Keywords Osteoarthritis; Artificial Intelligence; AI; Machine learning; CBCT; Cone Beam Computed Tomography; 3D; Panoramic
ID# 3543
Date of submission: 10/19/2023spacer
E-mail pewarchuk@livemail.uthscsa.edu
Author Cody Pewarchuk, DMD
Co-author(s) Joe Cordahi, DDS, MS, MBA
Co-author(s) e-mail cordahi@livemail.uthscsa.edu
Faculty mentor/Co-author Hassem Geha, DDS, MS
Faculty mentor/Co-author e-mail geha@uthscsa.edu
Basic Science Rationale
(Mechanisms that may account for and/or explain the clinical question, i.e. is the answer to the clinical question consistent with basic biological, physical and/or behavioral science principles, laws and research?)
post a rationale
None available
Comments and Evidence-Based Updates on the CAT
post a comment
None available

Return to Found CATs list