DOI: https://doi.org/10.22263/2312-4156.2024.4.94
O.N. Dudich1, V.S. Osipovich2, V.L. Krasilnikova1
The development of a software tool based on a convolutional neural network for processing multispiral computed tomography data of the orbit
1The Institute for Advanced Training & Retraining of Healthcare Personnel of the educational institution “Belarusian State Medical University”, Minsk, Republic of Belarus
2Belarusian State University of Informatics and Radioelectronics, Minsk, Republic of Belarus
Vestnik VGMU. 2024;23(4):94-104.
Abstract.
The article presents the results of research on the development and implementation in the form of a software tool for calculating the volume of the bone and soft tissue orbit, eye dystopia. To achieve this set goal, deep learning of the Mask R-CNN neural network was used. The defining moment in training the neural network is the definition of biomarkers – the main elements of the orbit, to which the neural network should pay its attention. The main biomarkers were: bone structures of the orbit, the eyeball, extraocular muscles and displaced retrobulbar tissue. The further capabilities of artificial intelligence to correctly assess and interpret them depend on the accuracy and correctness of the choice of biomarkers. It has been found that the error in calculating the volumes of the orbit based on the results of marking the neural network with the volumes of the orbits calculated based on the results of their manual marking does not exceed 8%. The developed software tool based on convolutional neural networks showed good results in the automatic calculation of the main anatomical and topographic parameters of the orbit and can be used in clinical practice when assessing the results of surgical intervention to reconstruct thin bones of the orbit.
Keywords: multispiral computed tomography, orbital fracture, 3D model, DICOM images, neural networks.
References
1. Levchenko OV, Krylov VV, Davydov DV, Lezhnev DA, Mikhaylyukov VM, Sharifullin FA, i dr. X-ray computed tomography to evaluate the effectiveness of surgical reconstruction of posttraumatic ocular defects and deformities. Neirokhirurgiya. 2014;(1):29-33. (In Russ.)
2. Levchenko OV, Mikhaylyukov VM, Davydov DV. Frameless navigation system for surgical treatment of posttraumatic defects and cranio-orbital deformities. Neirokhirurgiya. 2013;(3):9-14. (In Russ.). doi: http://dx.doi.org/10.17650/1683-3295-2013-0-3-9-14
3. Davydov DV, Lezhnev DA, Kostenko DI. MSCT diagnosis and planning of surgical treatment in traumas and posttraumatic deformations of ocular walls. DoktorRu. Gastroenterologiya. 2016;(1):116-20. (In Russ.)
4. Stuchilov VA, Nikitin AA. Optimization of diagnostics and surgical treatment of patients with fractures of the eye socket: posobie dlya vrachei. Moscow, RF; 2015. 36 р. (In Russ.)
5. Masalitina NN, Kurochka KS. Application of an automated classifier of computed tomography results to construct a geometric model of the human vertebrae. Dokl BGUIR. 2017;105(3):12-9. (In Russ.)
6. Rybalchenko GN; zayavitel' i patentoobladatel' Zakrytoe aktsioner o-vo «KONMET Inkorporeited». Rybalchenko plate for reconstruction of damaged bone structures of the zygomaticomandibular complex and method of reconstruction of damaged bone structures of the zygomaticomandibular complex with its use: pat 2202302 Ros Federatsiya: MPK A 61 B 17/80, A 61 B 17/56, A 61 F 2/28. № 2000125015/14; zayavl 04.10.00; opubl 20.04.03. (In Russ.)
7. Shalumov A-SZ. Method of manufacturing an individualized precision implant for replenishment of complex subtotal polyosseous ocular defect: pat 2164392 Ros Federatsiya: MPK A 61 F 2/28. № 2000116451/14; zayavl 27.06.00; opubl 27.03.01. (In Russ.)
8. The DICOM Standard is managed by the Medical Imaging & Technology Alliance – a division of the National Electrical Manufacturers Association [Internet]. Available from: https://www.dicomstandard.org/current/. [Accessed 21th August 2024].
9. Liu S, Qi L, Qin H, Shi J, Jia J. Path Aggregation Network for Instance Segmentation [Internet]. Available from: https://arxiv.org/pdf/1803.01534. [Accessed 21th August 2024].
10. Mask R-CNN – Inspect Weights of a Trained Model [Internet]. Available from: https://github.com/matterport/Mask_RCNN/blob/master/samples/coco/inspect_weights.ipynb. [Accessed 21th August 2024].
11. Karakulko AA, Medvedev OS, Moroz PA, Dudich ON, Krasilnikova VL, Osipovich VS, i dr. The use of neural networks in medical image processing. V: Bogush VA, i dr, redkol. BIG DATA and Advanced Analytics = BIG DATA i analiz vysokogo urovnya: sb materialov VI Mezhdunar nauch-prakt konf, Minsk, 20-21 maya 2020 g: v 3 ch. Ch 1. Minsk, RB: Bestprint; 2020. Р. 294-302. (In Russ.)
Submitted 20.06.2024
Accepted 28.08.2024
Information about authors:
O.N. Dudich – Candidate of Medical Sciences, associate professor of the Chair of Ophthalmology, the Institute for Advanced Training and Retraining of Healthcare Personnel of the Educational Institution “Belarusian State Medical University”, https://orcid.org/0009-0004-6554-3230,
e-mail: Этот адрес электронной почты защищён от спам-ботов. У вас должен быть включен JavaScript для просмотра. – Oksana N. Dudich;
V.S. Osipovich – Candidate of Technical Sciences, associate professor of the Chair of Engineering Psychology and Ergonomics, Belarusian State University of Informatics and Radioelectronics, https://orcid.org/0000-0001-9658-2866
V.L. Krasilnikova – Doctor of Medical Sciences, professor of the Chair of Ophthalmology, the Institute for Advanced Training and Retraining of Healthcare Personnel of the Educational Institution “Belarusian State Medical University”, https://orcid.org/0000-0002-5852-2616
