(46) AI-Enhanced Ultrasound in Pediatric Odontogenic Infections: Deep Learning for Diagnosing Facial Cellulitis Versus Abscess

Abstract: Ultrasound (US) is a fast, widely accessible, cost-effective, and pain-free diagnostic method that can be repeatedly used without posing any risk to the patient. The primary objective of this prospective study is to calibrate the efficacy of ultrasound (US) protocols in pediatric dental settings, particularly for diagnosing facial cellulitis and abscess. A secondary and crucial aim is to investigate the potential of developing deep learning AI algorithms to enhance the diagnostic capabilities of orofacial ultrasound in standard clinical use.
This research was conducted in a pediatric emergency service departments at two hospitals, Cincinnati Children’s Hospital Medical Center and Rambam Healthcare Campus, focusing on patients presenting with odontogenic facial swelling. The practitioners were trained in a specific US protocol using a high-frequency linear array transducer. AI models for use in POC ultrasound for the imaging and diagnosis of orofacial lesions using the data from Technion-Rambam and CCHMC Pediatric Dentistry patient cohort (n = 500). For this purpose, we trained and tested various deep CNN models (VGG, ResNet, Inception) on the existing cohort, along with using GAN-based models. This extensive collection of data was earmarked for the development and training of AI algorithms in deep learning environments, with the objective of enhancing precision in pediatric odontogenic infection diagnosis.
The results indicated a high concordance rate between the ultrasound evaluations conducted by trained practitioners, radiologists and trained AI. This suggests the effectiveness of ultrasound in diagnosing facial cellulitis and abscess among pediatric patients. Additionally, preliminary analysis of the collected ultrasound data demonstrated promising potential for the development of AI algorithms capable of enhancing diagnostic accuracy and efficiency in orofacial ultrasound.
The study furthers our understanding of potential for ultrasound protocol usage in pediatric dental practice, specifically for the diagnosis of facial cellulitis and abscess. It also highlights the transformative potential of integrating AI and deep learning into dental diagnostics. The development of AI algorithms based on deep learning could revolutionize the use of ultrasound in craniomaxillofacial surgery, offering more precise, efficient, and standardized diagnostic capabilities. This integration could significantly benefit patient outcomes by facilitating quicker and more accurate diagnoses, reducing unnecessary exposure to radiation, ultimately leading to more effective and timely treatments. The findings advocate further research and development in AI-enhanced diagnostic tools in clinical usage.

References:
1. Abdalla-Aslan R, Gaitini D, Shilo D, Emodi O, Beck-Razi N, Rachmiel A, Javitt M. Intraoral and transcutaneous point-of-care-ultrasound: an adjunct bedside tool for oral and maxillofacial clinicians. Quintessence Int. 2022 Mar 14;53(4):352-361. doi: 10.3290/j.qi.b2644877. PMID: 35119239.
2. Berger T, Garrido F, Green J, Lema PC, Gupta J. Bedside ultrasound performed by novices for the detection of abscess in ED patients with soft tissue infections. Am J Emerg Med. 2012 Oct;30(8):1569-73. doi: 10.1016/j.ajem.2011.08.002. Epub 2011 Oct 24. PMID: 22030180.
Workflow, train and test modelsCellulitis vs Abscess Ultrasound