Related Experiment Video
Updated: Oct 10, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
3.0K
Deep semi-supervised learning for automatic segmentation of inferior alveolar nerve using a convolutional neural
Ho-Kyung Lim1, Seok-Ki Jung2, Seung-Hyun Kim3
1Department of Oral and Maxillofacial Surgery, Korea University Guro Hospital, 148, Gurodong-ro, Guro-gu, Seoul, 08308, Republic of Korea.
BMC Oral Health
|December 8, 2021
Summary
Artificial intelligence (AI) was used to automatically segment the inferior alveolar nerve (IAN) in CT scans. While manual segmentation was more accurate, the AI approach significantly reduced segmentation time, offering a potentially faster surgical planning tool.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Surgical Planning
Background:
- The inferior alveolar nerve (IAN) is crucial for sensation in the lower jaw and lip.
- Accurate pre-surgical identification of the IAN's position is vital for patient safety.
- Current methods for IAN identification may lack speed and efficiency.
Purpose of the Study:
- To develop and evaluate an AI-driven method for automatic segmentation and tracking of the inferior alveolar nerve (IAN).
- To assess the accuracy and efficiency of the AI model compared to manual segmentation.
- To determine the potential of AI as a clinical tool for surgical planning.
Main Methods:
- Utilized 138 cone-beam computed tomography (CBCT) datasets from multiple centers.
- Employed a customized 3D nnU-Net architecture for image segmentation.
- Implemented an active learning framework with iterative dataset additions and evaluated accuracy using Dice Similarity Coefficient (DSC) and segmentation time.
Main Results:
- The Dice Similarity Coefficient (DSC) for IAN segmentation improved with active learning, reaching 0.58 ± 0.08.
- Segmentation time significantly decreased, with the final stage requiring only 86.4 seconds.
- Visual scoring indicated that manual segmentation achieved higher accuracy than the automated AI method.
Conclusions:
- A deep active learning framework demonstrates potential as a fast and robust tool for IAN demarcation.
- AI-assisted segmentation can expedite surgical planning by reducing the time required for IAN identification.
- Further refinement of AI models may be necessary to match the accuracy of manual segmentation in clinical practice.
