The purposes of the reseach
are to develop and justify a medical decision support system that uses an algorithm for classifying maxillary sinus pathologies in digital dia-fanoscopy based on the ResNet-50 convolutional neural network to increase the accuracy of detecting pathologies (sinusitis and cystic change) in the context of population health screening.
Methods
. The features of pathologies were considered, their forms were revealed, which are most often found in rhinological patients. A classifier of maxillary sinus pathologies was developed into three classes based on the results of the analysis of diaphanograms recorded using a digital diaphanoscopy software and hardware complex using the ResNet-50 convolutional neural network. The work proposes algorithms for a system for supporting medical decisionmaking.
Results
. A software tool has been developed for classifying the state of maxillary sinus tissues using the ResNet-50 convolutional neural network algorithm, which implements algorithms for learning and classifying maxillary sinus pathologies. The developed medical decision support system was tested. The following accuracy indicators were obtained when differentiating the condition of the maxillary sinuses into three classes: "absence of pathology" – sensitivity 0,94, specificity 0,96, accuracy 0,95; "sinusitis" – sensitivity 0,9, specificity 0,95 and accuracy 0,88; "cystic change" – sensitivity 0,86, specificity 0,96 and accuracy 0,88.
Conclusion
. The proposed medical decision support system provides more accurate diagnostics than existing economically available methods. In addition, the proposed solution is able to differentiate the types of pathological changes on diaphanograms in contrast to the previously proposed approaches in digital diaphanoscopy. In this way, the developed technology for optical diagnostics of the condition of the maxillary sinuses, together with the classification model developed on the basis of the SNS in the form of a system for supporting the adoption of medical decisions, can be used both as part of population health screening (for early detection of pathology), and as part of the diagnosis and control of the effectiveness of the prescribed therapy in otolaryngology, ENT oncology and dentistry, as well as in telemedicine.