Abstract

Topicality: Acute rhinosinusitis is an inflammatory disease affecting the nasal cavity and paranasal sinuses, with a duration of up to 12 weeks. While the frequency of post-viral rhinosinusitis is relatively low, acute viral rhinosinusitis can affect an individual 2 to 5 times a year. Unfortunately, acute rhinosinusitis often leads to unnecessary examinations and treatments. Although the routine use of x-ray methods for diagnosis is not recommended, sinus radiography is still commonly employed despite its limited diagnostic value. Therefore, enhancing the diagnostic efficiency of this method is crucial.

Objective: to evaluate the diagnosis of acute rhinosinusitis based on x-ray data using a customized neural network.

Results: A total of 900 x-rays of the sinuses were used to create a database, which was then divided into training, validation, and testing samples. The neural network was trained to perform binary classification on the samples. The accuracy of radiograph evaluation was as follows: 99.34% for the training set, 97.61% for the validation set, and 92.12% for the testing set, demonstrating high accuracy.

Hence, employing neural networks for recognizing normal and pathological conditions of the nasal cavity and paranasal sinuses is justified.

Conclusions: Based on our research, it is essential to develop and incorporate new criteria for evaluating sinus x-rays to enable the creation of neural networks with diagnostic accuracy exceeding 95%. The neural network proposed in this study can serve as an additional tool to enhance the accuracy of diagnosing acute rhinosinusitis using x-ray images, thereby promoting consensus among doctors regarding the identification of rhinosinusitis on radiographs and the need for further optional examinations or antibacterial agents.

Keywords: acute rhinosinusitis, rhinosinusitis, paranasal sinus x-ray, neural network, CNN, convolutional neural network.