Related Experiment Video
Updated: Jun 1, 2026
Automated Midline Shift and Intracranial Pressure Estimation based on Brain CT Images
Published on: April 13, 2013
Accuracy of a Cascade Network for Semi-Supervised Maxillary Sinus Detection and Sinus Cyst Classification
Xueqi Guo1, Zelun Huang2, Jieying Huang1
1Department of Oral Implantology, School and Hospital of Stomatology, Guangdong Engineering Research Center of Oral Restoration and Reconstruction & Guangzhou Key Laboratory of Basic and Applied Research of Oral Regenerative Medicine, Guangzhou Medical University, Guangzhou, Guangdong, China.
A new deep learning pipeline accurately detects and classifies maxillary sinus lesions from cone beam computed tomography (CBCT) scans. This AI tool aids surgical planning for maxillary sinus floor elevation by improving diagnostic precision.
Area of Science:
- Oral and Maxillofacial Radiology
- Artificial Intelligence in Medicine
- Medical Imaging Analysis
Background:
- Maxillary sinus mucosal cysts are common and require accurate diagnosis for effective surgical planning, particularly for maxillary sinus floor elevation procedures.
- Cone beam computed tomography (CBCT) is a key imaging modality for evaluating maxillary sinus pathologies.
Purpose of the Study:
- To develop and evaluate a deep learning (DL) pipeline for automated detection and classification of maxillary sinus lesions in CBCT images.
- To provide auxiliary diagnostic support for clinicians planning maxillary sinus floor elevation surgeries.
Main Methods:
- A cascade DL network was designed, integrating a semi-supervised object detection module for maxillary sinus area identification and a classification module for lesion categorization.
- The object detection module utilized a semi-supervised pseudo-labeling strategy to augment the training dataset, while Convolutional Neural Network (CNN) and Transformer architectures were compared for classification.
- Performance was assessed using Accuracy, Precision, Recall, F1 score, and Average Precision, with Grad-CAM used for visualization.
Main Results:
- The semi-supervised detection model achieved high accuracy (0.9403).
- ResNet-50 demonstrated superior classification performance with accuracies of 0.9836 (sagittal) and 0.9797 (coronal).
- Grad-CAM visualizations confirmed that the model focused on clinically relevant features for accurate lesion identification.
Conclusions:
- The developed DL pipeline offers high-precision detection and classification of maxillary sinus mucosal lesions.
- This automated approach reduces the need for extensive manual annotation while maintaining diagnostic accuracy.
- The pipeline serves as a valuable tool to support clinical decision-making in oral and maxillofacial surgery.
