Научная визуализация, 2026, том 18, номер 2, страницы 14 - 23, DOI: 10.26583/sv.18.2.02
Application of Semantic Segmentation Cascade Approach for Visualization of Optical Coherence Tomography Data
Авторы: V.V. Laptev1,A,B, V.V. Danilov2,A, E.A. Ovcharenko3,A, K.Yu. Klyshnikov4,A, I.S. Bessonov5,A, N.V. Litvinyuk6,A, N.A. Kochergin7,A
A Scientific Research Institute of Complex Problems of Cardiovascular Diseases, Kemerovo, Russia
B Siberian State Medical University, Tomsk, Russia
1 ORCID: 0000-0001-8639-8889, lptwlad1@gmail.com
2 ORCID: 0000-0002-1413-1381, viacheslav.v.danilov@gmail.com
3 ORCID: 0000-0001-7477-3979, ov.eugene@gmail.com
4 ORCID: 0000-0001-6247-1287, klyshnikovk@gmail.com
5 ORCID: 0000-0003-0578-5962, IvanBessnv@gmail.com
6 ORCID: 0000-0002-0630-7244, Nikita.litvinyuk@list.ru
7 ORCID: 0000-0002-1534-264X, nikotwin@mail.ru
Аннотация
A main goal in contemporary cardiology is to assess the risk of acute coronary syndrome (ACS) in individuals with ischemic heart disease in order to create preventative strategies and identify the best treatment plan. The objective of this research is to create an automated method for promptly identifying high-risk coronary lesions that are at risk of rupture (unstable plaques) in order to prevent ACS. We collected optical coherence tomography (OCT) volumes from 40 patients, with each OCT volume representing an RGB video of 704x704 pixels per frame, acquired over a certain depth. After filtering and manual annotation, 11,771 images were obtained to identify four types of objects: Lumen, Fibrous cap, Lipid core, and Vasa vasorum. To segment and quantitatively assess these features, we configured and evaluated the performance of three deep learning models (U-Net, MA-Net, DeepLabV3+). The study presents two approaches for training the aforementioned models: 1) detecting all analyzed objects and 2) applying a cascade of neural network models to separately detect subsets of objects. The results demonstrate the superiority of the cascade approach for analyzing OCT images. The combined use of DeepLabV3+ and MA-Net models achieved the highest average Dice similarity coefficient (DSC) of 0.747.
Ключевые слова: vascular segmentation, unstable plaques, optical coherence tomography, semantic segmentation, deep learning.
