Deep Learning in Diagnosis of Maxillary Sinusitis Using Conventional Radiography

Area of Science:

• Medical Imaging
• Artificial Intelligence in Radiology
• Deep Learning Applications

Background:

• Maxillary sinusitis is a common condition diagnosed using radiographic imaging.
• Waters' view radiographs are frequently employed for assessing paranasal sinuses.
• The diagnostic accuracy of human interpretation can vary, prompting the exploration of AI solutions.

Purpose of the Study:

• To evaluate the diagnostic efficacy of a deep learning (DL) algorithm for maxillary sinusitis detection.
• To compare the DL algorithm's performance against that of experienced radiologists.
• To assess diagnostic metrics including Area Under the Curve (AUC), sensitivity, and specificity.

Main Methods:

• A DL algorithm was developed using 8000 training and 1000 validation Waters' view radiographs.
• Two independent test sets (n=140 temporal, n=200 geographic) with CT correlation were used for evaluation.
• Performance was measured by AUC, sensitivity, specificity, and interobserver agreement (Cohen κ).

Main Results:

• The DL algorithm achieved significantly higher AUCs (0.93 temporal, 0.88 geographic) than radiologists (0.83-0.89 temporal, 0.75-0.84 geographic).
• Sensitivity and specificity of the DL algorithm were comparable to those of the radiologists.
• Strong interobserver agreement (Cohen κ = 0.82) and high correlation (0.89/0.84) were observed between the algorithm and radiologists.

Conclusions:

• The deep learning algorithm demonstrates superior diagnostic capability for maxillary sinusitis on Waters' view radiographs.
• The AI tool offers comparable sensitivity and specificity to human radiologists, indicating clinical potential.
• This AI approach shows promise for enhancing the accuracy and efficiency of sinusitis diagnosis.