The Use of Convolutional Neural Networks to Classify the States of the Maxillary Sinuses in Digital Diaphanoscopy

The work presents the results of the use of the convolutional neural network ResNet-50 in digital diaphanoscopy for the diagnosis of maxillary sinus conditions. The analysis of registered diaphanograms of patients with sinusitis, cystic fluid and conditionally healthy volunteers was carried out. It is shown that applying the proposed classification model to diaphanograms recorded at a sensing wavelength of 850 nm and fixation threshold of 80% allows to reduce the false negative result. The results analysis made it possible to establish requirements for the registered diaphanograms. An approach to dividing the developed model into static and dynamic components is proposed.