Quantitative evaluation of segmentation accuracy of subsegmental infarcts using 2DSTE and synthetic ultrasonic data in a spheroidal model of the left ventricle. (September 2022)

• Record Type: Journal Article
• Authors: Żmigrodzki, Jakub; Cygan, Szymon; Kałużyński, Krzysztof
• Is Part Of: Biomedical signal processing and control. Volume 78(2022)
• Publication Date: 2022-09
• Subjects: Block matching -- Echocardiography -- Infarct segmentation -- Speckle tracking echocardiography -- Strain imaging -- Synthetic ultrasonic data; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2022.103880
• Languages: English
• ISSNs: 1746-8094

Abstract

Abstract: Objective: Evaluation of the accuracy of the segmentation of simulated subsegmental infarcts using 2DSTE. Methods: Synthetic ultrasonic data were obtained using the FIELD II package and meshes obtained from FEM simulations of deformation of left ventricular (LV) models of a healthy LV and LVs with different infarcts. Hierarchical block matching method was applied to estimate displacements. Longitudinal and circumferential strain components were estimated and then averaged over small subsegments resulting from division of AHA 17 segments no 7 and 12. The infarcts were segmented in strain maps using three thresholds differing in sensitivity and specificity. The exact infarct location derived from FEM meshes and another one obtained from FEM derived displacement maps served as the references. The classification accuracy (ACC) and centroid position error (CPE) were used to assess the results of segmentation. Results: ACC does not drop below 0.92 and 0.85 when segmentation was done using error-free and 2DSTE estimated longitudinal strain maps, respectively. ACC does not drop below 0.90 and 0.75 in the same conditions for the circumferential strain maps. The mean CPE does not exceed 2 mm and 3 mm for infarct segmented from estimated longitudinal and circumferential strain maps, respectively. Conclusion: Reduction of the strain averaging area when using 2DSTE appeared necessary for detection and segmentation of small simulated infarcts (down to 9 mm diameter) as global strain measures showed insufficient sensitivity. Significance: The advantage of strain averaging over a reduced area for segmentation of small simulated infarcts when using the 2DSTE technique has been demonstrated.