SA1 - An MRI‑Guided and AI‑Assisted Approach to Reproducible TMJ Ultrasound Imaging

Abstract: The temporomandibular joint (TMJ) is essential for everyday function, enabling the coordinated movements required for mastication, speech, and facial expression. These functional demands predispose the joint to dysfunction and pain, often involving soft tissue pathology such as displacement or degeneration of the articular disc and joint capsule. Magnetic resonance imaging (MRI) is the gold standard for assessing TMJ soft tissues, however, its adoption in routine care is limited by high cost, prolonged scan times, patient discomfort, and access to hospital or imaging facilities. Cone beam computed tomography (CBCT) is more commonly available in dental clinics but is restricted to the evaluation of bony structures. In addition, neither MRI nor CBCT can capture dynamic TMJ function, which is important for identifying movement-dependent abnormalities such as disc displacement. As a result, many patients experience delayed or missed diagnoses, highlighting the need for an accessible, chairside imaging modality for TMJ soft tissue assessment.

Ultrasound represents a promising alternative due to its low cost, short acquisition time, lack of ionizing radiation, accessibility, and ability to provide real-time dynamic assessment. Despite these advantages, TMJ ultrasound is not widely adopted because of challenges related to anatomical interpretation, operator dependent acquisition, and limited reproducibility. The objective of this study is to develop a standardized TMJ ultrasound dataset with MRI-guided anatomical labeling and to integrate an AI-assisted TMJ localization system to improve the accuracy, reproducibility, and clinical usability of chairside TMJ ultrasound.

This translational pilot study collected paired 3T MRI and bilateral TMJ ultrasound data from 2 healthy adult participants (1 male and 1 female) under closed and open mouth positions. Dynamic ultrasound videos were obtained during maximum mouth opening and closing. MRI based segmentation of key anatomical structures, such as the condyle, articular disc, and joint capsule, was used as anatomical ground truth to guide labeling of the corresponding ultrasound images. All labels were independently reviewed by experienced musculoskeletal radiologists to ensure anatomical consistency. To assess the feasibility of object detection models for TMJ localization, a deep learning algorithm based on YOLO architecture was trained to identify the TMJ region using a publicly available dataset of 100 dental panoramic X-ray images from Roboflow Universe. Model performance was evaluated using mean average precision at an intersection-over-union threshold of 0.50.

MRI guided segmentation enabled accurate cross modality labeling of ultrasound images, with anatomical consistency confirmed by expert radiologist review. As a proof of concept, a YOLO-based object detection model achieved 100% mAP@50 on an 80-image training dataset and 98% mAP@50 on a 20-image test set. The resulting dataset establishes a clear link between TMJ ultrasound appearance and MRI defined anatomy across multiple probe orientations and functional states. In addition, dynamic ultrasound allowed measurement of condylar range of motion, enabling functional assessment that is not achievable with static imaging modalities. The AI-assisted localization framework demonstrated high accuracy in identifying the TMJ region, indicating potential to reduce operator dependence and improve acquisition consistency. Collectively, these outcomes demonstrate the feasibility of standardized, reproducible TMJ ultrasound imaging.

This study demonstrates that MRI-guided anatomical labeling of TMJ ultrasound images combined with an AI-assisted localization system can support consistent ultrasound image acquisition. Together, these approaches address key barriers to TMJ ultrasound adoption and establish a foundation for accessible, reproducible, chairside TMJ soft tissue imaging to improve temporomandibular disorder diagnosis and treatment.


References:

1. National Academies of Sciences, Engineering, and Medicine. Temporomandibular disorders: priorities for research and care. Washington, DC: National Academies Press; 2020. doi:10.17226/25652.

2. Maranini B, et al. The role of ultrasound in temporomandibular joint disorders: an update and future perspectives. Front Med (Lausanne). 2022;9:926573. doi:10.3389/fmed.2022.926573.