Deep Learning for Diagnosis of Paranasal Sinusitis Using Multi-View Radiographs

Area of Science:

• Radiology
• Artificial Intelligence
• Medical Imaging

Background:

• Interpreting Waters' and Caldwell view radiographs for sinusitis screening presents diagnostic challenges.
• Accurate sinusitis diagnosis is crucial for effective patient management and treatment.

Purpose of the Study:

• To develop and evaluate a deep learning algorithm for diagnosing frontal, ethmoid, and maxillary sinusitis.
• To compare the diagnostic performance of the algorithm against radiologists using both single and multi-view radiographic data.

Main Methods:

• A deep learning algorithm was trained and validated on 1403 radiographs, with a separate test set of 132 images.
• The algorithm processed both Waters' and Caldwell views simultaneously without manual cropping for sinus detection and classification.
• Area Under the Curve (AUC) was used to assess diagnostic performance, with statistical comparisons made against radiologist performance.

Main Results:

• The algorithm achieved satisfactory diagnostic performance for frontal (AUC=0.71), ethmoid (AUC=0.78), and maxillary sinusitis (AUC=0.88).
• The deep learning algorithm demonstrated higher AUC than radiologists for ethmoid and maxillary sinusitis (p=0.012 and p=0.013, respectively).
• A multi-view model outperformed a single Waters' view model for maxillary sinusitis diagnosis (p=0.038).

Conclusions:

• The developed deep learning algorithm exhibits diagnostic performance comparable to that of experienced radiologists.
• This AI-powered tool enhances the utility of radiography as a primary imaging modality for assessing multiple paranasal sinus infections.