THE PRACTICE OF USING ARTIFICIAL NEURAL NETWORKS TO RECOGNIZE IMAGES

(PATTERNS) OF THE TMJ STATE DURING CBCT

 

Chibisova M.A.^1,3, Kotina E.D.⁴, Pavlov A.V.², Korchagin D.S.⁴, Kuyumchian S.Yu.³,

Pridvizhkina T.S.³, Ermolaeva L.A.²

 

1 – I.I. Mechnikov North-Western State Medical University.

2 –Department of Therapeutic, Dentistry Medical Institute, St. Petersburg State University.

3 – N.I. Pirogov high-tech medical clinic, St. Petersburg State University.

4 – Faculty of applied mathematics – operation procedure, St. Petersburg State University. Saint-Petersburg, Russia.

P

urpose. To perform segmentation, geometry analysis and patterns of bone structures of the temporomandibular joint (TMJ) based on the data of cone beam computed tomography (CBCT), using an artificial neural network.

Materials and methods. In this work, we used CBCT data with TMJ capture in the open and closed position of the joint, DICOM data processing, file extraction and creation of masks to capture patterns (images) of TMJ geometry, density of cortical and spongy layers of the joint, and the formation of a database of masks of the articular process of the mandible in order to further recognize new input data by the neural network.

Results and discussion. According to the results of the quantitative assessment, the segmentation of the articular head was performed with the metrics Dice = 0.99 and IoU = 0.98. For the articular cavity, the metric values were Dice = 0.79 and IoU = 0.68. The differences between the results in the training and validation samples are insignificant, which is confirmed by similar metric values and the absence of pronounced readjustment of the model to the training data. The visual analysis of segmented masks corresponds to quantitative indicators and demonstrated stability of the results in various images. The data obtained allows us to proceed to the next stage – the extraction and analysis of the geometric characteristics of the identified structures.

The analysis of segmented masks of the articular head (condylar process of the mandibular branch) and the articular cavity of the temporal bone allows calculation of geometric parameters of diagnostic significance and used to assess the morphological state of the temporomandibular joint. These parameters include the height of the articular head of the mandibular branch and the articular fossa of the temporal bone, the anterior and posterior articular angles, the width of the articular fissure, and other indicators of linear angular measurements of the TMJ. These characteristics are important indicators of the condition of the temporomandibular joint, since changes in them may indicate various diseases.

Conclusion. The diagnostic potential of artificial neural networks as artificial intelligence technologies is undoubtedly an important and additional element for diagnostics. Since the interpretation and reading of CBCT data for a dentist and the description of CBT for a radiologist require increased attention and qualifications, respectively, in order to avoid diagnostic errors, in most cases an additional "second opinion" of artificial intelligence is needed. However, with the introduction of new technologies, specialists need not only to master them, but also to control the correctness of machine vision answers, so the final diagnosis and conclusion is made only by a doctor.

 

Keywords: artificial neural network, pattern, temporomandibular joint, TMJ, cone beam computed tomography, 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., Kotina E.D., Pavlov A.V., Korchagin D.S., Kuyumchian S.Yu., Pridvizhkina T.S., Ermolaeva L.A. The practice of using artificial neural networks to recognize images (patterns) of the TMJ state during CBCT. REJR 2025; 15(4):63-74. DOI: 10.21569/2222-7415-2025-15-4-63-74.

Received:        10.02.25                 Accepted:       26.11.25