Digital diaphanoscopy of maxillary sinus pathologies supported by machine learning.

Area of Science:

• Otorhinolaryngology
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Maxillary sinus pathologies are common ENT diseases needing prompt diagnosis.
• Traditional ENT inspection methods have limited sensitivity for detecting these conditions.
• Digital diaphanoscopy offers a novel approach for visualizing sinus alterations.

Purpose of the Study:

• To evaluate the efficacy of digital diaphanoscopy combined with machine learning for detecting maxillary sinus pathologies.
• To compare the performance of convolutional neural networks (CNNs) and linear discriminant analysis (LDA) for this diagnostic task.
• To assess the potential of these AI-driven methods as screening tools for sinus diseases.

Main Methods:

• Digital diaphanoscopy was employed to capture images of the maxillary sinuses.
• Two machine learning algorithms, CNNs and LDA, were applied to analyze the diaphanoscopy data.
• Performance metrics including sensitivity and specificity were calculated for each method.

Main Results:

• Both CNNs and LDA demonstrated higher sensitivity and specificity than traditional screening methods.
• Linear discriminant analysis achieved a sensitivity of 0.88 and a specificity of 0.98.
• LDA proved to be a simpler yet highly effective approach compared to CNNs for this application.

Conclusions:

• Digital diaphanoscopy integrated with machine learning, particularly LDA, shows significant promise for the accurate screening of maxillary sinus pathologies.
• This AI-enhanced diagnostic approach offers improved detection rates over conventional ENT examination techniques.
• The findings suggest a potential role for these technologies in improving early diagnosis and management of sinus conditions.