Related Experiment Video
Updated: Jan 18, 2026
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
Evaluation of deep learning-based segmentation models for carotid artery calcification detection in panoramic
Büşra Gül Yılmaz1, Sinan Altun2, İbrahim Şevki Bayrakdar3
1Department of Oral and Maxillofacial Radiology, University of Health Sciences, Mekteb-i Tıbbiye-i Şahane (Hamidiye) Complex, Selimiye Neighborhood, Tıbbiye Street, Üsküdar, Istanbul, 34668, Turkey. bsrglylmz@gmail.com.
Objectives:
The primary objective of this study is to evaluate the effectiveness of artificial intelligence-assisted segmentation methods in detecting carotid artery calcification (CAC) in panoramic radiographs and to compare the performance of different YOLO models: YOLOv5x-seg, YOLOv8x-seg, and YOLOv11x-seg. Additionally, the study aims to investigate the association between patient gender and the presence of CAC, as part of a broader epidemiological analysis.
Methods:
In this study, 30,883 panoramic radiographs were scanned. Annotations were made on 652 radiographs exhibiting features consistent with CAC, totaling 1,086 annotations. Deep learning-based analysis was conducted using three distinct YOLO segmentation models. The performance of these models was assessed using metrics such as precision, accuracy, and F1 score.
Results:
The YOLOv5x-seg model exhibited a balanced performance with a precision, sensitivity, and F1 score of 84.62% each. The YOLOv8x-seg model demonstrated higher sensitivity at 88.46%, albeit with a slightly higher false positive rate, evidenced by a precision of 78.63%. The YOLOv11x-seg model achieved the highest precision at 93.41%, an F1 score of 87.18%, and a sensitivity of 81.73%.
Conclusions:
AI-based segmentation models utilizing YOLO algorithms can be considered reliable tools for detecting CAC in panoramic radiographs. These models display promising performance appropriate for clinical applications; however, further research with larger and more diverse datasets is required to verify their generalizability and efficacy.
