Application of convolutional neural networks for the determination of maxillary sinus pathologies in digital diaphanoscopy
E.O. Bryanskaya,1 A.V. Kornaev,1 V.V. Dremin,1,2 Yu.O. Nikolaeva,3 V.G. Pil'nikov,3 A.V. Bakotina,3 A.V. Dunaev,1,
1 Orel State University named after I.S. Turgenev, Komsomolskaya St. 95, Orel, Russia
2 College of Engineering and Physical Sciences, Aston University, Birmingham B4 7ET, UK
3 A.I. Yevdokimov Moscow State University of Medicine and Dentistry, Delegatskaya St. 20, p. 1, Moscow 127473, Russia
Abstract
The digital diaphanoscopy method, which is based on optical probing of the sinuses and registration of scattering patterns of light, allows one to determine maxillary sinuses pathological changes. During the experimental studies, 55 conditionally healthy volunteers and 27 patients with maxillary sinuses pathologies were exanimated. The obtained results were compared with results of CT and MRI studies.
To differentiate the condition of the maxillary sinuses (presence or absence of pathology) the pattern recognition theory was used. The ResNet34 network was used as a classification model. The results obtained showed that the use of two wavelengths of probing the maxillary sinuses (650 and 850 nm) makes it possible to register light scattering patterns and differentiate of maxillary sinuses conditions into two classes (healthy and with pathological changes) with the following accuracy indicators: sensitivity – more than 70% and specificity – more than 90%.
These indicators exceed the values typical for the rhinoscopy method. According to the data from the literature, the values of rhinoscopy sensitivity and specificity for the detection of various pathologies of the maxillary sinuses are equal to 21-69% and 66-80%, respectively.
Thus, the use of convolutional neural networks to detect maxillary sinuses pathologies during digital diaphanoscopy has prospects for the diagnosis of patients, and also for the screening of the population.
The reported study was funded by RFBR according to the research project № 20-32-90147.
File with abstract
Speaker
Ekaterina Bryanskaya
Orel State University named after I.S. Turgenev
Russia
Discussion
Ask question
