A Multimodal Classification Method for Nasal Obstruction Severity Based on Computed Tomography and Nasal Resistance
Qiang Wang1, Shihao Li2, Hongzan Sun1
1Shengjing Hospital of China Medical University, Shenyang, China.
Annals of the New York Academy of Sciences
|October 3, 2025
Summary
This study introduces a novel multimodal model for classifying nasal obstruction severity using cone beam CT scans and nasal resistance data. The advanced model achieves high accuracy in diagnosing nasal obstruction, improving patient quality of life assessments.
Area of Science:
- Medical Imaging
- Computational Biology
- Otolaryngology
Background:
- Assessing nasal obstruction is crucial for diagnosing conditions and improving quality of life.
- Current methods may lack comprehensive analysis of obstruction severity.
Purpose of the Study:
- To develop a multimodal classification model for nasal obstruction degree.
- To integrate cone beam computed tomography (CBCT) imaging and nasal resistance measurements for enhanced accuracy.
Main Methods:
- A four-module model was designed: image feature extraction (MedicalNet, 3D CNN), table feature extraction (XGBoost), feature fusion (local and global), and classification (TabNet).
- MedicalNet was used for pre-training parameters, migrated to a 3D CNN for image analysis.
- Extreme gradient boosting (XGBoost) identified key features from nasal resistance data, reducing dimensionality.
- A novel feature fusion method combined local and global data characteristics.
Main Results:
- The multimodal model achieved high performance metrics.
- Accuracy reached 0.93, and recall reached 0.9.
- Experimental results demonstrated superior performance compared to existing methods.
Conclusions:
- The proposed multimodal classification model effectively assesses nasal obstruction degree.
- Integration of CBCT imaging and nasal resistance data offers significant diagnostic advantages.
- This approach holds promise for improved clinical diagnosis and patient management.
