APPLICATION OF ARTIFICIAL NEURAL NETWORKS IN THE RECOGNITION
OF X-RAY IMAGES IN OUTPATIENT DENTISTRY
Chibisova M.A.1,3, Ermolaeva L.A.2, Ilyin F.Yu.2, Pavlov A.V.2,
Mashroor MD Islam4, Binsumait Hashem M.A.4
1 – I.I. Mechnikov NWSMU.
2 – St. Petersburg State University, Medical Institute, Department of Therapeutic Dentistry.
3 – N.I. Pirogov HMTC, St. Petersburg State University.
4 – Faculty of Mathematics and Mechanics, St. Petersburg State University. Saint-Petersburg, Russia.
|
P |
urpose. The present study evaluates the diagnostic potential of artificial neural networks in interpreting X-ray data in the context of outpatient dentistry.
Materials and Methods. The study used 204 orthopantomographic images, 232 radiovisiograms, each of which contained at least three teeth, regardless of their location. These images were randomly divided into three datasets: 1 – training, 2 – validation during model development, 3 – testing. One neural network model was selected for the study, characterized by a fixed number of floating-point operations per second FLOPs(B) 42.6. The model was trained according to three scenarios with different epochs: 50, 100 and 200. Intermediate versions of the trained model were recorded every 50 epochs to evaluate the progression.
Results. The results emphasize the effectiveness of ANN in detecting and interpreting radiopaque structures on dental orthopantomograms and radiovisiograms, achieving an object detection accuracy of 0.98, provided that the training dataset contained sufficiently labeled objects. Increasing the number of training epochs increased both the accuracy of object detection (up to 0.875) and the quality of segmentation (up to 0.947), achieving optimal performance at 200 epochs. A comparative analysis showed that although dentists demonstrated 1% higher detection quality, the neural network processed images about 25 times faster. Errors in network forecasts mainly occurred due to insufficient training data for certain categories of objects, which confirms the need for a reliable and well-chosen training dataset.
Discussion. The analysis of radiological research methods using artificial neural network technology opens up a new horizon for dentists and radiologists, not only for automated recognition of the pathology of hard tissues of teeth and jaws, but also for automated formation of a second opinion and routing of patients to specialists in related fields.
Conclusion. The use of artificial neural networks for the detection and interpretation of X-ray images in outpatient dentistry demonstrates significant diagnostic potential. Neural network models optimized for object recognition and segmentation provide a high level of reliability. However, the quality of the results depends on the availability of a sufficiently diverse and representative training dataset. This study highlights the importance of comprehensive dataset preparation to maximize the accuracy of detection and delineation of radiographic features.
Keywords: artificial neural network, object detection, orthopantomogram, RVG, CBCT.
Corresponding author: Pavlov A.V., e-mail: Этот e-mail адрес защищен от спам-ботов, для его просмотра у Вас должен быть включен Javascript
For citation: Chibisova M.A., Ermolaeva L.A., Ilyin F.Yu., Pavlov A.V.,Mashrour MD Islam, Binsumait Hashem M.A. The use of artificial neural networks in the recognition of X-ray image data in outpatient dentistry. REJR 2025; 15(1):41-61. DOI: 10.21569/2222-7415-2025-15-1-41-61.
Received: 30.12.24 Accepted: 04.03.25