Related Experiment Video
Updated: Jun 27, 2025
05:56
Objectification of Tongue Diagnosis in Traditional Medicine, Data Analysis, and Study Application
Published on: April 14, 2023
2.5K
Design and evaluation of a deep learning-based automatic segmentation of maxillary and mandibular substructures using
L Melerowitz1, S Sreenivasa1, M Nachbar1
1Charité - Universitätsmedizin Berlin, Corporate Member of Freie Universität Berlin and Humboldt-Universität zu Berlin, Department of Radiation Oncology, Augustenburger Platz 1, 13353, Berlin, Germany.
Clinical and Translational Radiation Oncology
|May 7, 2024
Summary
This study introduces an automated method for segmenting jaw substructures in head and neck cancer patients, significantly reducing segmentation time and improving accuracy for radiation therapy planning.
Area of Science:
- Medical imaging and radiation oncology.
- Artificial intelligence in healthcare.
- Computational anatomy and segmentation.
Background:
- Current radiation treatment planning for head and neck cancer patients (HNCP) often treats the entire mandible as an organ at risk, neglecting detailed maxillary segmentation.
- Accurate risk assessment for osteoradionecrosis (ORN) and dental implant success requires precise dose analysis in specific mandibular and maxillary segments.
- Manual segmentation of jaw structures is time-consuming, inconsistent, and lacks standardized definitions for subsections.
Purpose of the Study:
- To develop and validate an automated segmentation model for 12 distinct mandibular and maxillary substructures.
- To assess the efficiency and accuracy of the automated model compared to manual segmentation.
- To evaluate the model's performance in the presence of metal artifacts in CT scans.
Main Methods:
- A 3D U-Net deep learning model was trained on 82 CT scans from HNCP.
- The mandible and maxilla were divided into 12 substructures for segmentation.
- The automated method was validated against manual segmentation on 20 independent CT scans using Dice Similarity Coefficient (DSC), 95% Hausdorff Distance (HD95), and surface DSC (sDSC).
Main Results:
- Automated segmentation completed in a median of 86 seconds, drastically faster than the 53.5 minutes for manual segmentation.
- Median DSC ranged from 0.81 to 0.91, and median HD95 ranged from 1.61 to 4.22, indicating high accuracy.
- The model performed well even with metal artifacts, though maxillary substructures showed slightly lower segmentation metrics than mandibular ones.
Conclusions:
- The developed jaw substructure segmentation model is accurate, time-efficient, and robust, even with CT scans containing metal artifacts.
- This automated approach facilitates detailed dose analysis, crucial for predicting normal tissue complications like ORN and dental implant failure.
- The model offers a promising tool for advancing normal tissue complication probability (NTCP) models in radiation oncology.
