Automated assessment of nasal septum deviation using cone-beam computed tomography images based on artificial...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Orthodontics

Background:

• Nasal septum deviation (NSD) contributes to impaired nasal function and dentofacial abnormalities.
• Cone-beam computed tomography (CBCT) aids NSD diagnosis, but manual interpretation is time-consuming and requires expertise.

Purpose of the Study:

• To develop and evaluate an automated two-stage artificial intelligence (AI) framework for efficient and accurate NSD screening using CBCT scans.
• To assess the performance of different AI models in detecting and classifying NSD.

Main Methods:

• A dataset of 330 CBCT scans was used to train a two-stage AI framework.
• The first stage utilized the YOLOv11 object detection algorithm for nasal septum localization.
• The second stage employed convolutional neural networks (ResNet, EfficientNet, MobileNet) for NSD classification.

Main Results:

• YOLOv11n achieved high performance in nasal septum detection (precision 0.996, recall 1.000).
• Mobile_small demonstrated strong classification performance (AUC 0.817, accuracy 0.749).
• AI-assisted diagnosis improved orthodontists' accuracy by over 20% and reduced diagnosis time by 53.92%.

Conclusions:

• The proposed AI system enables rapid NSD screening with diagnostic accuracy comparable to manual interpretation.
• Lightweight AI models are viable for clinical CBCT analysis.
• AI-assisted diagnosis significantly enhances orthodontists' accuracy and efficiency in identifying NSD.