Deep-Learning-Based Automatic Segmentation of Parotid Gland on Computed Tomography Images

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Accurate segmentation of the parotid gland is crucial for head and neck imaging analysis.
• Manual segmentation is time-consuming and subject to inter-observer variability.

Purpose of the Study:

• To develop and evaluate a deep learning algorithm for automatic parotid gland segmentation on CT images.
• To assess the performance of a U-Net based model for this task.

Main Methods:

• Retrospective analysis of 30 head and neck CT volumes (931 axial images).
• Ground truth labeling by oral and maxillofacial radiologists using CranioCatch Annotation Tool.
• Development of a U-Net deep convolutional neural network model.
• Performance evaluation using F1-score, precision, sensitivity, and Area Under Curve (AUC).

Main Results:

• The AI model achieved an F1-score, precision, and sensitivity of 1 for parotid gland segmentation.
• The Area Under Curve (AUC) value was 0.96.
• Successful segmentation was defined as >50% pixel intersection with ground truth.

Conclusions:

• Deep learning models, specifically U-Net architecture, can effectively automate parotid gland segmentation on axial CT images.
• The developed AI model shows high accuracy and potential for clinical use in head and neck radiology.
• Automated segmentation offers a promising approach to improve efficiency and consistency in medical image analysis.