Evaluation of Segmentation Algorithms for the Detection of Plaque Border and Wall Cross-Sectional Area in B-Mode Intravascular Ultrasound Images

Abstract

The catheter-based diagnostic modality for an arterial system is an intravascular ultrasound (IVUS) image. This modality
provides the cross-sectional view of morphological information about the arterial wall including the lumen, media,
adventitia and also show the quantification of plaque components and wall thickness in two-dimensional (2D) image
format. The detection of lumen and media-adventitia borders in IVUS image constitutes a necessary step for quantitative
assessment of atherosclerotic lesions. Most of the segmentation methods reported are either manual or semi-automated,
requiring user interaction to some extent, which increases the analysis time and detection errors. In this work, a fully
automated approach for intimae and media-adventitia borders detection is proposed. The signal-dependent noise presence
is degraded the ultrasound imaging and artifacts lead to incorrect image measurements. This need to be removed and
smoothen by the homomorphic wavelet-based isotropic filter and also preserve the edge features. The segmentation and
borders detection of an intima (inner) and media-adventitia (outer) is performed by fuzzy-based region active contour. The
fuzzy-based region active contour method proposed is efficient in detecting Intimae and media-adventitia borders. These
are analyzed and compared with distance regularized level set evolution (DRLSE) and active contour method for IVUS
image. The performance of segmentation and border detecting methods are evaluated in terms of quantitative parameters
such as True Positive (TP), True Negative (TN), False Positive (FP) and False Negative (FN). The metric measurement of
accuracy, sensitivity, specificity, precision and recall values are computed for both manual and automatic border detection
methods. The contour metric measurement such as Jaccard Coefficient (JC), Missed detection (MD), Hausdorff Distance
(HD) and Percentage of Area Difference (PAD) are computed and visual variation of these performances.