Blood Detection in IVUS Images for 3D Volume of Lumen Changes Measurement Due to Different Drugs Administration

Abstract

Lumen volume variations is of great interest by the physicians given it reduces the probability of infarction as it increases. In this paper we present a fast and efficient method to detect the lumen borders in longitudinal cuts of IVUS sequences using an AdaBoost classifier trained with several local features assuring their stability. We propose a criterion for feature selection based on stability leave-one-out cross validation. Results on the segmentation of 18 IVUS pullbacks show that the proposed procedure is fast and robust leading to 90% of time reduction with the same characterization performance.

ECOC Random Fields for Lumen Segmentation in Radial Artery IVUS Sequences

“…Features are first extracted from the whole image, then a feature vector x i is assigned to each node S i ⊂ S by considering the median value of each feature in the block. The feature selection process proposed in [5] has been used to select, among a wide set of texture descriptors, the most discriminant features for the proposed problem, resulting in Gabor filters [9], Local Binary Patterns [10], Sobel filter in thex direction, mean value, standard deviation and the ratio among these two values on the grey-levels of the image, computed by a sliding windows of size H. Finally, the First Order Absolute Moment (FOAM) of the grey levels [11] is also used. FOAM is an operator that computes a vector always pointing towards the strongest gray-level discontinuity and assuming magnitude close to zero when applied to the discontinuity itself.…”

“…In [4] a shape space is constructed from statistical analysis of morphology on training data and borders are constrained to a smooth closed geometry. In [5] the lumen detection is achieved by classifying blood areas using Adaboost.…”

Blood flow imaging of high frame rate two‐dimensional vector in cardiovascular ultrasound detection

“…Based on the SVM method, the user is required to provide the first frame of the lumen and non‐inner region of the IVUS sequence to generate an initial likelihood value [18]. Rotger et al [19] used IVUS image grayscale and gradient information, Gabor and local binary pattern (LBP) features and AdaBoost classifier to segment the image to obtain the target edge. Yan et al [20] first extracted dense features for each pixel of the image, then used ECOC to obtain likelihood estimation, and then semantically segmented the IVUS image, and finally detected the film edge with the Snake model Based on the block, Su et al [21] combined the double sparse self‐encoder neural network to identify the intima and film edge points of the IVUS image.…”

Classification of blood regions in IVUS images using three dimensional brushlet expansions

“…In addition, intrinsic challenges associated with IVUS data such as the presence calcified plaques, bifurcations, guide wires, and more importantly motion of the catheter as well as the heart make existing algorithms only partially successful in clinical applications. As an alternative approach, several research teams have strived to detect blood regions or reduce blood noise effects in IVUS images, which could potentially be utilized as a preprocessing step for the detection of true lumen borders [6,7]. The authors in [8] also presented a 3D supervised classification approach (one-class support vector machine) using three spatial, one temporal and three frequency-based features.…”