Carotid Artery Segmentation in Ultrasound Images and Measurement of Intima-Media Thickness

Table 1

Overview of recent CCA IMC segmentation techniques for ultrasound imaging.

Name,; year, and; ref. no.	Common carotid artery IMT segmentation technique	Advantages & limitations of the methods	Selection method of ROI; (SA/FA)	Performance metric	Processing; time/frame 𝑇avg	IMT error; (mm)	𝑁
Ilea et al., 2013; [20]	Model-based approach, video tracking procedure: spatially coherent algorithm	Advantages: Method can deal with inconsistencies in the appearance of the IMC over the cardiac cycle. Robustness with respect to data captured under different imaging conditions.	FA	MAD	80 sec; (8 sec in 1st frame & 72 sec in tracking 28 frames)	(−0.007) ± 0.176	40*
Xu et al.,; 2012,; [21]	Hough transform and dual snake model	Advantages: It is less likely to be affected by noise, compensates the holes or missing boundaries, and can estimate the missing LI interface boundary.; Limitations: The method would work fine for early thickening of IMC but fails for irregular boundaries in the presence of plaques and eliminates minor details.	SA	MAD	0.465 sec	0.02 ± 0.03	50
Molinari et al.,; 2012,; [22]	Multiresolution edge snapper	Advantages: Complete automation, robustness to noise, and real-time computation.; Limitations: Robustness with respect to noise, but the LI/MA representation is less accurate.	FA	MAD	Less than 15 sec	0.078 ± 0.112	365
Destrempes et al.,; 2011,; [23]	Nakagami distributions, Bayesian model	Advantages: Robust to a reasonable variability in the initialization, lowest tracing error for LI & MA, method is not sensitive to the degree of stenosis or calcification.; Limitations: Depends on initial segmentation	SA	MAD; HD	38 sec	—	8988
Petroudi et al., 2011, [24]	Active contours & active contours without edges	Advantages: Fully automated, fast, does not require any user interaction, and works well for noisy images.	FA	MAD	—	0.09 ± 0.10	30
Destrempes et al.,; 2009,; [25]	Nakagami distributions, stochastic optimization	Advantages: Reasonable average computation time, robust to the estimation procedures.; Limitations: Method suitable for healthy arteries, extensive tuning & training, so computational cost is high, for different scanner requires retraining & retuning.	SA	MAD; HD	24 sec	—	7283
Faita et al.,; 2008, [26]	First-order absolute moment edge operator	Advantages: Suited for fast real-time implementation, operator can have immediate feedback on the quality of the images.; Limitations: Depends on ROI selection.	SA	MAD	—	0.001 ± 0.035	150
Liang et al.,; 2000,; [27]	Multiscale dynamic programming	Advantages: No initial human setting, capable of processing images of different quality, ambiguous cases user can intervene, and reduced interobserver variability.; Limitations: Training required, for different scanner retraining needed, searched LI & MA interfaces may cross each other.	FA	MAD	0.7 min	0.042 ± 0.02	50
Gustavsson et al., 1994,; [28]	Dynamic programming	Advantages: Fully automated, low computational complexity; suitable for clinical purposes, human correction allowed.; Limitations: Initial human setting & training required, fails for slanting IMC with weak boundary.	SA	MAD	—	0.03 ± 0.032	22

𝑁: number of images/cases, SA: semiautomated, FA: fully automated, SD: standard deviation, HD: Hausdorff distance, MAD: mean absolute distance, *video sequences, 𝑇_avg: average processing time/frame or image.