Fully-automated Alveolar Bone Level Measurements in Adolescents via Landmark Localization in Intraoral Ultrasound Videos

• Published in: 2023 IEEE International Ultrasonics Symposium (IUS) pp. 1 - 4
• Main Authors: Kumaralingam, Logiraj, Dinh, Hoang B. V., Nguyen, Kim-Cuong T., Punithakumar, Kumaradevan, Kaipatur, Neelambar R., Lou, Edmond H. M., Major, Paul W., Le, Lawrence H.
• Format: Conference Proceeding
• Language: English
• Published: IEEE 03.09.2023
• Subjects: Alveolar bone level; Bones; Convolutional neural networks; Deep learning; Landmark localization; Location awareness; Loss measurement; Malocclusion; Ultrasonic imaging; Ultrasonic variables measurement; Ultrasound videos; Uncertainty
• ISSN: 1948-5727
• DOI: 10.1109/IUS51837.2023.10308346

Accurate and automatic assessment of alveolar bone level in ultrasound videos is crucial for orthodontic treatment and diagnosis, as manual interpretation is time-consuming and clinicians exhibit substantial interobserver variation. A systematic approach for quantifying alveolar bone loss involves the direct measurements of the alveolar bone level (ABL), the distance between the cementoenamel junction and alveolar bone crest. In this paper, we propose an end-to-end landmark localizing network by combining a convolutional neural network and Swin-transformer architecture to effectively localize the cementoenamel junction and alveolar bone crest in intraoral ultrasound videos and automatically measure the ABL. In addition, key frames and non-key frames are identified and discarded based on an uncertainty prior. This study used ultrasound videos of 147 teeth acquired from 20 orthodontic adolescent patients. Experimental studies have been performed and compared with state-of-the-art models to prove the feasibility of the proposed architecture.