AI-based segmentation of the parotid gland on CT images

1. Title:

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

2. Authors: Önder M.; Evli C.; Türk E.; Kazan O.; Bayrakdar İ.Ş.; Çelik Ö.; Costa A.L.F.; Gomes J.P.P.; Ogawa C.M.; Jagtap R.; Orhan K.

Summary

This retrospective study aimed to develop and evaluate a deep convolutional neural network (dCNN) based on U-Net architecture for the automatic segmentation of parotid glands using axial computed tomography (CT) images. The researchers utilized 30 anonymized CT volumes, which were processed into 931 axial slices for training, validation, and testing. Ground truth labeling was performed by two experts using the CranioCatch Annotation Tool. The resulting AI model demonstrated exceptional performance, achieving an F1-score, precision, and sensitivity of 1.0, with an Area Under Curve (AUC) of 0.96. The findings suggest that deep learning can successfully automate the time-consuming process of manual organ segmentation in medical imaging

4. Key Words: Artificial intelligence; Deep learning; U-Net; Parotid gland; Computed tomography

5. Extracted data

5.1. Year: 2023

5.2. Modality: Computed Tomography (CT)

5.3. Dataset: 30 CT volumes; 931 axial CT slices

5.4. Dataset Split: Training: 745 (80%); Validation: 93 (10%); Testing: 93 (10%)

5.5. Network Architecture: U-Net based

5.6. Metrics: F1-score, Precision, Sensitivity, AUC

5.7. AP - Professional Qty: 2

5.8. AP - Supervisor Presence: No information

5.9. AP - Experience Level: 11 years and 2 years

5.10. AP - Expertise Area: Oral and Maxillofacial Radiology

5.11. AP - Tool or System: CranioCatch Annotation Tool

5.12. ML Task: Semantic Segmentation

5.13. Project Objective: Automatic segmentation of the parotid gland on CT images using deep learning

6. Clinical Relevance: Automatic segmentation tools can remove the burden of repetitive tasks for radiologists, such as manual organ or nerve segmentation, allowing clinicians to focus on complex clinical issues. In the context of the parotid gland, precise imaging and segmentation assist in clinical decision-making regarding the management of malignant or benign tumors, which require different surgical and therapeutic approaches. Additionally, these AI tools can enhance computer-aided detection, treatment planning, and dental student education by providing consistent and reliable anatomical models. The high accuracy of this model suggests it can minimize human error and maximize clinical efficiency