Related Experiment Video
Updated: May 30, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
2.6K
Assessment of deep learning technique for fully automated mandibular segmentation
Ebru Yurdakurban1, Yağızalp Süküt2, Gökhan Serhat Duran3
1Department of Orthodontics, Faculty of Dentistry, Muğla Sıtkı Koçman University, Muğla, Turkey.
Summary
An open-source AI model accurately segments mandibles from CT scans, showing high similarity to manual segmentation. This tool is suitable for clinical use, enabling custom model development by clinicians.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Radiology
Background:
- Mandible segmentation is crucial for dental and maxillofacial diagnostics.
- Automated segmentation can improve efficiency and consistency in clinical workflows.
Purpose of the Study:
- To evaluate the precision of an open-source, clinician-trained convolutional neural network (CNN) model for automated mandible segmentation.
- To assess the model's performance against manual segmentation using various metrics.
Main Methods:
- 55 cone-beam computed tomography (CBCT) scans were used for training and testing.
- The MONAI Label tool facilitated the training of the CNN model.
- Performance was evaluated using Dice similarity coefficient, Hausdorff distance, precision, recall, and volumetric/surface deviation analysis.
Main Results:
- The automated model achieved a high Dice similarity coefficient (0.926 ± 0.014).
- Mean precision and recall values were 0.941, with minimal surface and volumetric differences compared to manual segmentation.
- No statistically significant differences were found in surface or volumetric comparisons.
Conclusions:
- The automated mandible segmentation model is precise and suitable for clinical application.
- Open-source software empowers clinicians to develop tailored automated segmentation solutions.
