Enhancing airway obstruction diagnosis with multimodal 3D shape analysis

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Otolaryngology

Background:

• Enlarged adenoids obstruct nasal breathing, leading to health issues like cognitive deficits and developmental delays.
• Current diagnostic methods (polysomnography, visual inspection) are often inaccurate, time-consuming, or costly.
• Cone-beam computed tomography (CBCT) scans are common for patients with suspected adenoid hypertrophy.

Purpose of the Study:

• Develop an open-source, automated deep learning tool for quantitative assessment of airway obstruction.
• Utilize CBCT scans to automatically segment and extract 3D airway morphology for diagnosis.
• Improve diagnostic accuracy and efficiency for enlarged adenoids.

Main Methods:

• Employ a deep learning approach combining multi-view and point cloud representations for 3D shape analysis.
• Process CBCT scans to capture both global and local airway features.
• Develop a tool for automated segmentation and quantitative airway obstruction assessment.

Main Results:

• Achieved 81.88% accuracy in classifying the presence or absence of adenoid hypertrophy.
• Demonstrated improved performance in predicting the nasopharynx airway obstruction ratio.
• Model shows promise in detecting severe cases, with ongoing refinement for all severity levels.

Conclusions:

• The automated tool offers rapid, quantitative, and reproducible airway obstruction assessments.
• Potential to significantly enhance clinical workflows and diagnostic efficiency.
• Promising solution for improving patient outcomes in the diagnosis of enlarged adenoids.