DWT-based segmentation method for coronary arteries

Towards automated coronary artery segmentation: A systematic review

BACKGROUND AND OBJECTIVE

Vessel segmentation is the first processing stage of 3D medical images for both clinical and research use. Current segmentation methods are tedious and time consuming, requiring significant manual correction and hence are infeasible to use in large data sets.

METHODS

Here, we review and analyse available coronary artery segmentation methods, focusing on fully automated methods capable of handling the rapidly growing medical images available. All manuscripts published since 2010 are systematically reviewed, categorised into different groups based on the approach taken, and characteristics of the different approaches as well as trends over the past decade are explored.

RESULTS

The manuscripts were divided intro three broad categories, consisting of region growing, voxelwise prediction and partitioning approaches. The most common approach overall was region growing, particularly using active contour models, however these have had a sharp fall in popularity in recent years with convolutional neural networks becoming significantly more popular.

CONCLUSIONS

The systematic review of current coronary artery segmentation methods shows interesting trends, with rising popularity of machine learning methods, a focus on efficient methods, and falling popularity of computationally expensive processing steps such as vesselness and multiplanar reformation.

A three-dimensional quantification of calcified and non-calcified plaques in coronary arteries based on computed tomography coronary angiography images: Comparison with expert's annotations and virtual histology intravascular ultrasound

The detection, quantification and characterization of coronary atherosclerotic plaques has a major effect on the diagnosis and treatment of coronary artery disease (CAD). Different studies have reported and evaluated the noninvasive ability of Computed Tomography Coronary Angiography (CTCA) to identify coronary plaque features. The identification of calcified plaques (CP) and non-calcified plaques (NCP) using CTCA has been extensively studied in cardiovascular research. However, NCP detection remains a challenging problem in CTCA imaging, due to the similar intensity values of NCP compared to the perivascular tissue, which surrounds the vasculature. In this work, we present a novel methodology for the identification of the plaque burden of the coronary artery and the volumetric quantification of CP and NCP utilizing CTCA images and we compare the findings with virtual histology intravascular ultrasound (VH-IVUS) and manual expert's annotations. Bland-Altman analyses were employed to assess the agreement between the presented methodology and VH-IVUS. The assessment of the plaque volume, the lesion length and the plaque area in 18 coronary lesions indicated excellent correlation with VH-IVUS. More specifically, for the CP lesions the correlation of plaque volume, lesion length and plaque area was 0.93, 0.84 and 0.85, respectively, whereas the correlation of plaque volume, lesion length and plaque area for the NCP lesions was 0.92, 0.95 and 0.81, respectively. In addition to this, the segmentation of the lumen, CP and NCP in 1350 CTCA slices indicated that the mean value of DICE coefficient is 0.72, 0.7 and 0.62, whereas the mean HD value is 1.95, 1.74 and 1.95, for the lumen, CP and NCP, respectively.

An Automatic 3-D Reconstruction of Coronary Arteries by Stereopsis

Stereopsis of X-ray images can produce 3D tree of coronary arteries up to a certain accuracy level with a lower dose of radiation when compared to computer tomography (CT). In this study, a novel and complete automatic system is designed that covers preprocessing, segmentation, matching and reconstruction steps for that purpose. First, an automatic and novel pattern recognition technique is applied for extraction of the bifurcation points with their diameters recorded in a map. Then, a novel optimization algorithm is run for matching the branches efficiently which is based on that map and the epipolar geometry of stereopsis. Finally, cut branches are fixed one by one at the bifurcations for completing the 3D reconstruction. This method prevails the similar ones in the literature with this novelty since it automatically and inherently prevents the wrong overlapping of branches. Other essential problems like correct detection of the bifurcations and accurate calibration parameters and fast overlapping of matched branches are addressed at acceptable levels. The accuracy of bifurcation extraction is high at 90 % with 96 % sensitivity. Accuracy of vessel centerlines has rootmean-square (rms) error smaller than 0.57 mm for 20 different patients. For phantom model, rms error is 0.75 ± 0.8 mm in 3D localization.

Effect of Watermarking on Diagnostic Preservation of Atherosclerotic Ultrasound Video in Stroke Telemedicine

Embedding of diagnostic and health care information requires secure encryption and watermarking. This research paper presents a comprehensive study for the behavior of some well established watermarking algorithms in frequency domain for the preservation of stroke-based diagnostic parameters. Two different sets of watermarking algorithms namely: two correlation-based (binary logo hiding) and two singular value decomposition (SVD)-based (gray logo hiding) watermarking algorithms are used for embedding ownership logo. The diagnostic parameters in atherosclerotic plaque ultrasound video are namely: (a) bulb identification and recognition which consists of identifying the bulb edge points in far and near carotid walls; (b) carotid bulb diameter; and (c) carotid lumen thickness all along the carotid artery. The tested data set consists of carotid atherosclerotic movies taken under IRB protocol from University of Indiana Hospital, USA-AtheroPoint™ (Roseville, CA, USA) joint pilot study. ROC (receiver operating characteristic) analysis was performed on the bulb detection process that showed an accuracy and sensitivity of 100 % each, respectively. The diagnostic preservation (DPsystem) for SVD-based approach was above 99 % with PSNR (Peak signal-to-noise ratio) above 41, ensuring the retention of diagnostic parameter devalorization as an effect of watermarking. Thus, the fully automated proposed system proved to be an efficient method for watermarking the atherosclerotic ultrasound video for stroke application.

Analysis of enlarged images using time-of-flight magnetic resonance angiography, computed tomography, and conventional angiography

This study aimed to assess the accuracy of time-of-flight magnetic resonance angiography, computed tomography, and conventional angiography in depicting the actual length of the blood vessels. Three-dimensional time-of-flight magnetic resonance angiography and computed tomography angiography were performed using a flow phantom model that was 2.11 mm in diameter and had a total area of 0.26 cm(2). After this, volume rendering technique and the maximum intensity projection method as well as two-dimensional digital subtraction angiography and three-dimensional rotational angiography based on conventional angiography were conducted. For three-dimensional time-of-flight magnetic resonance angiography, 8 channel sensitivity encoding (SENSE) head coil for the 3.0 Tesla equipment was used. Fluid was added to the normal saline solution at various rates, such as 11.4, 20.0, 31.4, 40.0, 51.5, 60.0, 71.5, 80.1, 91.5, and 100.1 cm/s using an automatic contrast media injector. Each image was thoroughly examined. After reconstructing the image using the maximum intensity projection method, the length of the conduit in the center of the coronal plane was measured 30 times. After performing computed tomography angiography with the 64-channel CT scanner and 16-channel CT scanner, the images were sent to TeraRecon. Then, the length of the conduit in the center of the coronal plane of each image was measured 30 times after reconstructing the images using volume rendering and maximum intensity projection techniques. For conventional angiography, three-dimensional rotational angiography and two-dimensional digital subtraction angiography were used. Images obtained by three-dimensional rotational angiography were reconstructed and enhanced by 33, 50, and 100 % in the 128 Matrix and the 256 Matrix, respectively on the Xtra Vision workstation. The maximum intensity projection was used for the reconstruction, and the length of the conduit was measured 30 times in the center of the coronal plane of each image. Measurements using the two-dimensional digital subtraction angiography were obtained 30 times in the center of the image. As a result, the lumen length measured by three-dimensional enhanced flow MR angiography images was a minimum of 2.51 ± 0.12 mm when the fluid velocity was 40 cm/s. The images obtained by computed tomography angiography were larger than the actual images obtained by using the test equipment and the reconstruction method. Among the reconstruction methods of three-dimensional rotational angiography, the lumen length in the image reconstructed by 100 % in the 256 matrix was the smallest; 2.76 ± 0.009 mm. In the 128 matrix, as the scope of reconstruction was widened, the length of the vessel was increased by 0.710 units. In the 256 matrix, as the scope of reconstruction was widened, the length of the vessel was decreased by 0.972 units. In two-dimensional digital subtraction angiography, the lumen length in the image was 2.22 ± 0.095 mm. Although this image was magnified similar to the image reconstructed by 100 % in the 256 matrix of three-dimensional rotational angiography (P < 0.05), it was closest to the actual image among the images compared in this study. In conclusion, images obtained by two-dimensional digital subtraction angiography were closer to the actual images compared to the images obtained by other procedures. It can be concluded that vascular images obtained by magnetic resonance angiography, CT angiography, and conventional angiography were larger than the actual images.

Brain MR image segmentation with spatial constrained K-mean algorithm and dual-tree complex wavelet transform

In brain MR images, the noise and low-contrast significantly deteriorate the segmentation results. In this paper, we propose an automatic unsupervised segmentation method integrating dual-tree complex wavelet transform (DT-CWT) with K-mean algorithm for brain MR image. Firstly, a multi-dimensional feature vector is constructed based on the intensity, the low-frequency subband of DT-CWT and spatial position information. Then, a spatial constrained K-mean algorithm is presented as the segmentation system. The proposed method is validated by extensive experiments using both simulated and real T1-weighted MR images, and compared with the state-of-the-art algorithms.