2026
Brattain, Jose Gabriel Gonzalez Nunez; Melody Halbert; Matthew Lewis; Brian Kirkwood; Kevin Gillespie; Claudia Maydana; Laura
A Unified Evaluation of Tooth Segmentation Models and Diffusion-Based Synthetic Data Augmentation in Panoramic Dental Radiographs Conference
IEEE-EMBS International Conference on Body Sensor Networks (BSN 2026), 2026.
Abstract | Tags: BSN, REU | Links:
@conference{nokey,
title = {A Unified Evaluation of Tooth Segmentation Models and Diffusion-Based Synthetic Data Augmentation in Panoramic Dental Radiographs},
author = { Jose Gabriel Gonzalez Nunez; Melody Halbert; Matthew Lewis; Brian Kirkwood; Kevin Gillespie; Claudia Maydana; Laura Brattain},
url = {https://openreview.net/forum?id=MRHsDIo4DT},
year = {2026},
date = {2026-10-10},
urldate = {2026-10-10},
publisher = { IEEE-EMBS International Conference on Body Sensor Networks (BSN 2026)},
abstract = {Panoramic dental radiographs (PDRs) are among the most commonly used imaging modalities in dentistry, providing comprehensive visualization of dental and maxillofacial structures for preventive care and pathology assessment. Interpretation of these images, however, relies on manual examination, which can be challenged by image complexity and clinical workload. Subtle abnormalities and early pathological changes may escape detection, highlighting the need for advanced computational tools. Recent advances in artificial intelligence (AI) have enabled more accurate and detailed interpretation of diagnostic imaging, yet the development of robust AI systems for dental imaging remains limited by the scarcity of large, diverse, and AI-ready annotated datasets. We systematically evaluated state-of-the-art (SOTA) AI architectures and foundation models for tooth segmentation in PDRs within a unified evaluation framework. We compared two YOLO-based segmentation architectures (YOLOv8m and YOLO11m) with two boxprompted Segment Anything Model (SAM) variants, MobileSAM and EfficientViT-SAM, under identical training and evaluation conditions. YOLO11m achieved the best performance, achieving a Dice coefficient of 0.90, IoU of 0.82, and Recall of 0.95, substantially outperforming the SAM-based models (Dice 0.66–0.67). To mitigate the data limitation, we further investigated synthetic image generation through a class-conditional denoising diffusion probabilistic model (DDPM), trained to generate individual teeth. The model achieved an Frechet Inception Distance (FID) of ´ 87.9 and an Inception Score of 3.33. Targeted augmentation of underrepresented tooth classes, particularly third molars and adjacent second molars, produced a consistent positive trend in macro-F1 score across five independent training runs. These results demonstrate the potential of diffusion-generated synthetic data to mitigate class imbalance in dental AI applications.},
keywords = {BSN, REU},
pubstate = {published},
tppubtype = {conference}
}
Panoramic dental radiographs (PDRs) are among the most commonly used imaging modalities in dentistry, providing comprehensive visualization of dental and maxillofacial structures for preventive care and pathology assessment. Interpretation of these images, however, relies on manual examination, which can be challenged by image complexity and clinical workload. Subtle abnormalities and early pathological changes may escape detection, highlighting the need for advanced computational tools. Recent advances in artificial intelligence (AI) have enabled more accurate and detailed interpretation of diagnostic imaging, yet the development of robust AI systems for dental imaging remains limited by the scarcity of large, diverse, and AI-ready annotated datasets. We systematically evaluated state-of-the-art (SOTA) AI architectures and foundation models for tooth segmentation in PDRs within a unified evaluation framework. We compared two YOLO-based segmentation architectures (YOLOv8m and YOLO11m) with two boxprompted Segment Anything Model (SAM) variants, MobileSAM and EfficientViT-SAM, under identical training and evaluation conditions. YOLO11m achieved the best performance, achieving a Dice coefficient of 0.90, IoU of 0.82, and Recall of 0.95, substantially outperforming the SAM-based models (Dice 0.66–0.67). To mitigate the data limitation, we further investigated synthetic image generation through a class-conditional denoising diffusion probabilistic model (DDPM), trained to generate individual teeth. The model achieved an Frechet Inception Distance (FID) of ´ 87.9 and an Inception Score of 3.33. Targeted augmentation of underrepresented tooth classes, particularly third molars and adjacent second molars, produced a consistent positive trend in macro-F1 score across five independent training runs. These results demonstrate the potential of diffusion-generated synthetic data to mitigate class imbalance in dental AI applications.
2024
Khodagholi, Zahra; Sun, Jasmine; Awad, Nour; Vankayalapati, Amulya; Dion, Gregory R.; Brattain, Laura J.
Ultrasound Landmark Tracking for Real-Time Anatomcial Airway Identification and Reliability Check Conference
2024 IEEE 20th International Conference on Body Sensor Networks (BSN), 2024.
Abstract | Tags: BSN, REU | Links:
@conference{Khodagholi2024,
title = {Ultrasound Landmark Tracking for Real-Time Anatomcial Airway Identification and Reliability Check},
author = {Zahra Khodagholi and Jasmine Sun and Nour Awad and Amulya Vankayalapati and Gregory R. Dion and Laura J. Brattain},
url = {https://ieeexplore.ieee.org/document/10780557},
doi = {https://doi.org/10.1109/BSN63547.2024.10780557},
year = {2024},
date = {2024-10-15},
urldate = {2024-10-15},
publisher = {2024 IEEE 20th International Conference on Body Sensor Networks (BSN)},
abstract = {Ultrasound has become an essential tool for airway management, providing non-invasive, real-time imaging that enhances the procedure efficacy. The main challenges of ultrasound include the specialized training needed for interpretation and the high patient variability. This paper presents ULTRA-AIR (Ultrasound Landmark Tracking for Real-time Anatomical Airway Identification and Reliability Check), a deep learning system designed for real-time tracking of multiple neck anatomical landmarks to assist with ultrasound-guided airway management. Uncertainty estimation was also computed to quantify model confidence scores, providing safety measures for trustworthy AI in real-time. Utilizing the YOLOv9 architecture, ULTRA-AIR was fine-tuned using Leave-One-Out cross-validation on neck ultrasound data collected from six human subjects, total of 3,060 images and 11,142 annotated bounding boxes. The system achieved AUCs of 0.99, 0.95, 0.89, and 0.77 for the classes of strap muscle, thyroid gland, tracheal ring, and thyroid cartilage, respectively. Average F1 score was 0.83 and mAP50 was 0.84. High uncertainty scores helped to identify cases needing human intervention. ULTRA-AIR shows promise for safe, accurate, and rapid airway assessments at point of care.},
keywords = {BSN, REU},
pubstate = {published},
tppubtype = {conference}
}
Ultrasound has become an essential tool for airway management, providing non-invasive, real-time imaging that enhances the procedure efficacy. The main challenges of ultrasound include the specialized training needed for interpretation and the high patient variability. This paper presents ULTRA-AIR (Ultrasound Landmark Tracking for Real-time Anatomical Airway Identification and Reliability Check), a deep learning system designed for real-time tracking of multiple neck anatomical landmarks to assist with ultrasound-guided airway management. Uncertainty estimation was also computed to quantify model confidence scores, providing safety measures for trustworthy AI in real-time. Utilizing the YOLOv9 architecture, ULTRA-AIR was fine-tuned using Leave-One-Out cross-validation on neck ultrasound data collected from six human subjects, total of 3,060 images and 11,142 annotated bounding boxes. The system achieved AUCs of 0.99, 0.95, 0.89, and 0.77 for the classes of strap muscle, thyroid gland, tracheal ring, and thyroid cartilage, respectively. Average F1 score was 0.83 and mAP50 was 0.84. High uncertainty scores helped to identify cases needing human intervention. ULTRA-AIR shows promise for safe, accurate, and rapid airway assessments at point of care.