Catheter detection and segmentation in volumetric ultrasound using SVM and GLCM; Научная визуализация; Т. 10, № 4

Catheter detection and segmentation in volumetric ultrasound using SVM and GLCM; Научная визуализация; Т. 10, № 4

Bibliographic Details
Parent link:Научная визуализация: электронный журнал.— , 2009-
Т. 10, № 4.— 2018.— [С. 30-39]
Corporate Author: Национальный исследовательский Томский политехнический университет Инженерная школа информационных технологий и робототехники Отделение информационных технологий
Other Authors: Danilov V. V. Vyacheslav Vladimirovich, Skirnevsky I. P. Igor Petrovich, Manakov R. Roman, Kolpashchikov D. Yu. Dmitry Yurjevich, Gerget O. M. Olga Mikhailovna, Malgani F.
Summary:Заглавие с экрана
The focus of this study was to develop an image-based algorithm for the catheter detection and segmentation in volumetric ultrasound. Nowadays, echocardiography is one of the most common methods of cardiovascular diseases diagnostic and surgery. As an input data the algorithm uses epicardial full-volume 3D echocardiography datasets. In total, 9 datasets consisted of 15 three-dimensional timeframes were processed. Each 3D timeframe includes 208 slices with the size of 176*176. To correctly detect the catheter, the feature-based approach was applied to recognition the catheter within the 3D echocardiography datasets. MATLAB was used for all calculations as the main numerical computing environment. Before the main part of the algorithm, we performed pre-processing of the data. The pre-processing workflow comprises imposing a restriction on the area of the region for noise reduction, automatic Otsu’s thresholding and morphological operations. The proposed algorithm based on gray-level co-occurrence matrix (GLCM) was applied as a feature extraction technique. Once the GLCM was computed, we obtained correlation, contrast, homogeneity and energy features. Then we applied feature thresholds to the catheter detection. These thresholds were obtained using Support Vector Machine (SVM) with the linear kernel function and standardization the predictor data. The average segmentation and recognition accuracies of the algorithm equal 94.16% and 87.2% respectively. The processing time for one 2D slice and one 3D dataset are equal to 9±0.2 milliseconds and 1.96±0.045 seconds, respectively. Though the algorithm is not time-consuming for 2D mode, it is still complicated to apply it to 3D real-time visualization.
Language:English
Published: 2018
Subjects:
электронный ресурс
труды учёных ТПУ
catheter detection
catheter segmentation
texture analysis
GLCM
ultrasound
Online Access:http://dx.doi.org/10.26583/sv.10.4.03
Format: Electronic Book Chapter
KOHA link:https://koha.lib.tpu.ru/cgi-bin/koha/opac-detail.pl?biblionumber=659378

Similar Items

Similar Items