Dice Metric Research Articles

Sparse group composition for robust left ventricular epicardium segmentation

Left ventricular (LV) epicardium segmentation in cardiac magnetic resonance images (MRIs) is still a challenging task, where the a-priori knowledge like those that incorporate the heart shape model is usually used to derive reasonable segmentation results. In this paper, we propose a sparse group composition (SGC) approach to model multiple shapes simultaneously, which extends conventional sparsity-based single shape prior modeling to incorporate a-priori spatial constraint information among multiple shapes on-the-fly. Multiple interrelated shapes (shapes of epi- and endo-cardium of myocardium in the case of LV epicardium segmentation) are regarded as a group, and sparse linear composition of training groups is computed to approximate the input group. A framework of iterative procedure of refinement based on SGC and segmentation based on deformation model is utilized for LV epicardium segmentation, in which an improved shape-constraint gradient Chan-Vese model (GCV) acted as deformation model. Compared with the standard sparsity-based single shape prior modeling, the refinement procedure has strong robust for relative gross and not much sparse errors in the input shape and the initial epicardium location can be estimated without complicated landmark detection due to modeling spatial constraint information among multiple shapes effectively. Proposed method was validated on 45 cardiac cine-MR clinical datasets and the results were compared with expert contours. The average perpendicular distance (APD) error of contours is 1.50±0.29mm, and the dice metric (DM) is 0.96±0.01. Compared to the state-of-the-art methods, our proposed approach appealed competitive segmentation performance and improved robustness.

Localization and Segmentation of 3D Intervertebral Discs in MR Images by Data Driven Estimation.

This paper addresses the problem of fully-automatic localization and segmentation of 3D intervertebral discs (IVDs) from MR images. Our method contains two steps, where we first localize the center of each IVD, and then segment IVDs by classifying image pixels around each disc center as foreground (disc) or background. The disc localization is done by estimating the image displacements from a set of randomly sampled 3D image patches to the disc center. The image displacements are estimated by jointly optimizing the training and test displacement values in a data-driven way, where we take into consideration both the training data and the geometric constraint on the test image. After the disc centers are localized, we segment the discs by classifying image pixels around disc centers as background or foreground. The classification is done in a similar data-driven approach as we used for localization, but in this segmentation case we are aiming to estimate the foreground/background probability of each pixel instead of the image displacements. In addition, an extra neighborhood smooth constraint is introduced to enforce the local smoothness of the label field. Our method is validated on 3D T2-weighted turbo spin echo MR images of 35 patients from two different studies. Experiments show that compared to state of the art, our method achieves better or comparable results. Specifically, we achieve for localization a mean error of 1.6-2.0 mm, and for segmentation a mean Dice metric of 85%-88% and a mean surface distance of 1.3-1.4 mm.

Automatic segmentation of the left ventricle in cardiac MRI using local binary fitting model and dynamic programming techniques.

Segmentation of the left ventricle is very important to quantitatively analyze global and regional cardiac function from magnetic resonance. The aim of this study is to develop a novel algorithm for segmenting left ventricle on short-axis cardiac magnetic resonance images (MRI) to improve the performance of computer-aided diagnosis (CAD) systems. In this research, an automatic segmentation method for left ventricle is proposed on the basis of local binary fitting (LBF) model and dynamic programming techniques. The validation experiments are performed on a pool of data sets of 45 cases. For both endo- and epi-cardial contours of our results, percentage of good contours is about 93.5%, the average perpendicular distance are about 2 mm. The overlapping dice metric is about 0.91. The regression and determination coefficient between the experts and our proposed method on the LV mass is 1.038 and 0.9033, respectively; they are 1.076 and 0.9386 for ejection fraction (EF). The proposed segmentation method shows the better performance and has great potential in improving the accuracy of computer-aided diagnosis systems in cardiovascular diseases.

Multi-atlas segmentation with augmented features for cardiac MR images.

Multi-atlas segmentation infers the target image segmentation by combining prior anatomical knowledge encoded in multiple atlases. It has been quite successfully applied to medical image segmentation in the recent years, resulting in highly accurate and robust segmentation for many anatomical structures. However, to guide the label fusion process, most existing multi-atlas segmentation methods only utilise the intensity information within a small patch during the label fusion process and may neglect other useful information such as gradient and contextual information (the appearance of surrounding regions). This paper proposes to combine the intensity, gradient and contextual information into an augmented feature vector and incorporate it into multi-atlas segmentation. Also, it explores the alternative to the K nearest neighbour (KNN) classifier in performing multi-atlas label fusion, by using the support vector machine (SVM) for label fusion instead. Experimental results on a short-axis cardiac MR data set of 83 subjects have demonstrated that the accuracy of multi-atlas segmentation can be significantly improved by using the augmented feature vector. The mean Dice metric of the proposed segmentation framework is 0.81 for the left ventricular myocardium on this data set, compared to 0.79 given by the conventional multi-atlas patch-based segmentation (Coupé et al., 2011; Rousseau et al., 2011). A major contribution of this paper is that it demonstrates that the performance of non-local patch-based segmentation can be improved by using augmented features.

TH-C-18A-02: Machine Learning and STAPLE Based Simultaneous Longitudinal Segmentation of Bone and Marrow Structures From Dual Energy CT

Purpose: To develop a fully-automatic longitudinal bone and marrow segmentation method in the pelvic region from dual energy computed tomography (DECT). Methods: We developed a two-step automatic bone and marrow segmentation method for simultaneous longitudinal evaluation of patients with metastatic bone disease using dual energy CT (DECT). Our approach transforms the DECT images into a multi-material decomposition (MMD) model that represents the voxels as a mixture of multiple materials. A support vector machine (SVM) was trained using a single scan. In the first step of the longitudinal segmentation the trained SVM model detects bone and marrow structures on all available longitudinal scans. Segmentation is further refined through active contour segmentation. In the second step, the segmentations from the individual scans are merged by employing the simultaneous truth and performance level estimation (STAPLE) algorithm. The scans are registered using affine and deformable registration. We found that our approach improves the segmentation in all the scans under reliable registration performance between the same scans. Improving registration was not under the scope of this work. Results: We applied our approach to segment bone and marrow in DECT scans in the pelvic regions for multiple patients. Each patient had three to five follow upmore » scans. All the patients in the analysis had artificial metal prostheses which introduced challenges for the registration. Our algorithm achieved reasonable accurate segmentation despite the presence of metal artifacts and high-density oral contrast in neighboring structures. Our approach obtained an overall segmentation accuracy of 80% using DICE metric. Conclusion: We developed a two-step automatic longitudinal segmentation technique for bone and marrow region structures in the pelvic areas from dual energy CT. Our approach achieves robust segmentation despite the presence of confounding structures with similar intensities as the bone and marrow regions due to uptake of oral contrast and metal artifacts.« less

Is Hamming distance only way for matching binary image feature descriptors?

Brute force matching of binary image feature descriptors is conventionally performed using the Hamming distance. This paper assesses the use of alternative metrics in order to see whether they can produce feature correspondences that yield more accurate homography matrices. Two statistical tests, namely ANOVA (Analysis of Variance) and McNemar's test were employed for evaluation. Results show that Jackard-Needham and Dice metrics can display better performance for some descriptors. Yet, these performance differences were not found to be statistically significant.

Fully Automated Ultrasound Common Carotid Artery Segmentation Using Active Shape Model

B-mode measurement of the contour of the common carotid artery (CCA) has an important clinical value. The purpose of this study was to develop a fully automated ultrasound common carotid artery segmentation method using active shape model (ASM). An image database with 90 images was used to train the ASM model during the offline training phase of ASM. When it came to the online segmentation phase, a knowledge-based seed point detection method was first used to locate the centroid of the CCA. Then the trained ASM model automatically produced an exact contour of the CCA. The proposed method yielded a Dice Metric of 90.5% ± 4.35% and a Hausdorff Distance of 9.28 ± 5.2 pixels in a database of 40 ultrasound images. The segmentation result of upper and bottom part of the CCA was better than that of lateral part of the CCA. The proposed method eliminates the need of manual initialization, and identifies the contour of the CCA with high precision. It has the potential to be a suitable replacement for manual segmentation of the CCA.

Endocardial Border Detection Using Radial Search and Domain Knowledge

The ejection fraction rate is a frequently used parameter when treating patients who suffered from heart disease. However, the measurement of this ejection rate depends on manual segmentation of left ventricle cavity in the end-systolic and end-diastolic phases. This paper proposes a semi-automatic algorithm for the detection of left ventricular border in two dimensional long axis ultrasound echocardiographic images. First, we apply a preprocessing filter to the ultrasound for the sake of speckle reduction. Then the knowledge of the anatomical structure of human heart and local homogeneity of blood pool is being used to detect the border of left ventricle. The proposed method evaluates 80 ultrasound images from four healthy volunteers and the generated contours are compared with contours manually drawn by an expert. The measured Dice Metric and Hausdorff Distance recorded by the proposed algorithm are 85.1% ± 0.4% and 3.25 ± 0.46 mm respectively. The numerical results reported in this paper indicate that the proposed algorithm is able to correctly segment the left ventricle cavity and can be used as an alternative to manual contouring of left ventricle cavity from ultrasound images.

Unsupervised tissue segmentation from dynamic contrast-enhanced magnetic resonance imaging

Design, implement, and validate an unsupervised method for tissue segmentation from dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI). For each DCE-MRI acquisition, after a spatial registration phase, the time-varying intensity of each voxel is represented as a sparse linear combination of adaptive basis signals. Both the basis signals and the sparse coefficients are learned by minimizing a functional consisting of a data fidelity term and a sparsity inducing penalty. Tissue segmentation is then obtained by applying a standard clustering algorithm to the computed representation. Quantitative estimates on two real data sets are presented. In the first case, the overlap with expert annotation measured with the DICE metric is nearly 90% and thus 5% more accurate than state-of-the-art techniques. In the second case, assessment of the correlation between quantitative scores, obtained by the proposed method against imagery manually annotated by two experts, achieved a Pearson coefficient of 0.83 and 0.87, and a Spearman coefficient of 0.83 and 0.71, respectively. The sparse representation of DCE MRI signals obtained by means of adaptive dictionary learning techniques appears to be well-suited for unsupervised tissue segmentation and applicable to different clinical contexts with little effort.

Prostate MRI Segmentation Using Learned Semantic Knowledge and Graph Cuts

We propose a fully automated method for prostate segmentation using random forests (RFs) and graph cuts. A volume of interest (VOI) is automatically selected using supervoxel segmentation, and its subsequent classification using image features and RF classifiers. The VOIs probability map is generated using image and context features, and a second set of RF classifiers. The negative log-likelihood of the probability maps acts as the penalty cost in a second-order Markov random field cost function. Semantic information from the second set of RF classifiers is an important measure of each feature to the classification task, which contributes to formulating the smoothness cost. The cost function is optimized using graph cuts to get the final segmentation of the prostate. With average dice metric (DM) (on the training set) and DM (on the test set), our experimental results show that inclusion of the context and semantic information contributes to higher segmentation accuracy than other methods.

Left Ventricle Segmentation in Magnetic Resonance Images with Modified Active Contour Method

Desired segmentation of the image is a pivotal problem in image processing. Segmenting the left ventricle (LV) in magnetic resonance images (MRIs) is essential for evaluation of cardiac function. Fo r the segmentation of cardiac M RI several methods have been proposed and implemented. Each of them has advantages and restrictions. A modified region-based active contour model was applied for segmentation of LV chamber. A new semi-automatic algorithm was suggested calculating the appropriate Balloon force according to mean intensity of the region of interest for each image. The database is included of 2,039 MR images collected fro m 18 children under 18. The results were co mpared with previous literatures according to two standards: Dice Metric (DM) and Point to Curve (P2C). The obtained segmentation results are better than previously reported values in several literatures. In this study different points were used in cardiac cycle and several slice levels and classified into three levels: Base, Mid. and Apex. The best results were obtained at end diastole (ED) in co mparison with end systole (ES), and on base slice than other slices, because of LV b igger size in ED phase and base slice. With segmentation of LV M RI based on novel active contour and application of the suggested algorithm for balloon force calculat ion, the mean imp rovement of DM co mpared to Grosgeorge et al. is 19.6% in ED and 49.5% in ES phase. The mean improvement of P2C co mpared with the same literature respectively for ED and ES phase is 43.8% and 39.6%.

Left ventricular segmentation from MRI datasets with edge modelling conditional random fields.

BackgroundThis paper considers automatic segmentation of the left cardiac ventricle in short axis magnetic resonance images. Various aspects, such as the presence of papillary muscles near the endocardium border, makes simple threshold based segmentation difficult.MethodsThe endo- and epicardium are modelled as two series of radii which are inter-related using features describing shape and motion. Image features are derived from edge information from human annotated images. The features are combined within a discriminatively trained Conditional Random Field (CRF). Loopy belief propagation is used to infer segmentations when an unsegmented video sequence is given. Powell’s method is applied to find CRF parameters by minimizing the difference between ground truth annotations and the inferred contours. We also describe how the endocardium centre points are calculated from a single human-provided centre point in the first frame, through minimization of frame alignment error.ResultsWe present and analyse the results of segmentation. The algorithm exhibits robustness against inclusion of the papillary muscles by integrating shape and motion information. Possible future improvements are identified.ConclusionsThe presented model integrates shape and motion information to segment the inner and outer contours in the presence of papillary muscles. On the Sunnybrook dataset we find an average Dice metric of 0.91±0.02 and 0.93±0.02 for the inner and outer segmentations, respectively. Particularly problematic are patients with hypertrophy where the blood pool disappears from view at end-systole.

Hybrid segmentation of left ventricle in cardiac MRI using gaussian-mixture model and region restricted dynamic programming

Segmentation of the left ventricle from cardiac magnetic resonance images (MRI) is very important to quantitatively analyze global and regional cardiac function. The aim of this study is to develop a novel and robust algorithm which can improve the accuracy of automatic left ventricle segmentation on short-axis cardiac MRI. The database used in this study consists of three data sets obtained from the Sunnybrook Health Sciences Centre. Each data set contains 15 cases (4 ischemic heart failures, 4 non-ischemic heart failures, 4 left ventricle (LV) hypertrophies and 3 normal cases). Three key techniques are developed in this segmentation algorithm: (1) ray scanning approach is designed for segmentation of images with left ventricular outflow tract (LVOT), (2) a region restricted technique is employed for epicardial contour extraction, and (3) an edge map with non-maxima gradient suppression approach is put forward to improve the dynamic programming to derive the epicardial boundary. The validation experiments were performed on a pool of data sets of 45 cases. For both endo- and epi-cardial contours of our results, percentage of good contours is about 91%, the average perpendicular distance is about 2mm. The overlapping dice metric is about 0.92. The regression and determination coefficient between the experts and our proposed method on the ejection fraction (EF) is 1.01 and 0.9375, respectively; they are 0.9 and 0.8245 for LV mass. The proposed segmentation method shows the better performance and is very promising in improving the accuracy of computer-aided diagnosis systems in cardiovascular diseases.

A region-based segmentation of tumour from brain CT images using nonlinear support vector machine classifier

The proposed system provides new textural information for segmenting tumours, efficiently and accurately and with less computational time, from benign and malignant tumour images, especially in smaller dimensions of tumour regions of computed tomography (CT) images. Region-based segmentation of tumour from brain CT image data is an important but time-consuming task performed manually by medical experts. The objective of this work is to segment brain tumour from CT images using combined grey and texture features with new edge features and nonlinear support vector machine (SVM) classifier. The selected optimal features are used to model and train the nonlinear SVM classifier to segment the tumour from computed tomography images and the segmentation accuracies are evaluated for each slice of the tumour image. The method is applied on real data of 80 benign, malignant tumour images. The results are compared with the radiologist labelled ground truth. Quantitative analysis between ground truth and the segmented tumour is presented in terms of segmentation accuracy and the overlap similarity measure dice metric. From the analysis and performance measures such as segmentation accuracy and dice metric, it is inferred that better segmentation accuracy and higher dice metric are achieved with the normalized cut segmentation method than with the fuzzy c-means clustering method.

Automatic Left Ventricle Segmentation in Cardiac MRI Using Topological Stable-State Thresholding and Region Restricted Dynamic Programming

Segmentation of the left ventricle (LV) is very important in the assessment of cardiac functional parameters. The aim of this study is to develop a novel and robust algorithm which can improve the accuracy of automatic LV segmentation on short-axis cardiac magnetic resonance images (MRI). The database used in this study consists of 45 cases obtained from the Sunnybrook Health Sciences Centre. The 45 cases contain 12 ischemic heart failures, 12 non-ischemic heart failures, 12 LV hypertrophies, and 9 normal cases. Three key techniques are developed in this segmentation algorithm: 1) topological stable-state thresholding method is proposed to refine the endocardial contour, 2) an edge map with non-maxima gradient suppression approach, and 3) a region-restricted technique that is proposed to improve the dynamic programming to derive the epicardial boundary. The validation experiments were performed on a pool of data sets of 45 cases. For both endo- and epicardial contours of our results, percentage of good contours is about 91%, the average perpendicular distance is about 2 mm, and the overlapping dice metric is about 0.91. The regression and determination coefficient for the experts and our proposed method on the ejection fraction is 1.05 and 0.9048, respectively; they are 0.98 and 0.8221 for LV mass. An automatic method using topological stable-state thresholding and region restricted dynamic programming has been proposed to segment left ventricle in short-axis cardiac MRI. Evaluation results indicate that the proposed segmentation method can improve the accuracy and robust of left ventricle segmentation. The proposed segmentation approach shows the better performance and has great potential in improving the accuracy of computer-aided diagnosis systems in cardiovascular diseases.

SU‐E‐T‐25: A Learning Method for Assessing Margin Definition from Daily Image Deformations

Purpose: To investigate a new tool for quantification of deformations and optimal margin estimation from daily images using learning algorithms based on swarm intelligence. Methods: Swarm intelligence methods, specifically Ant Colony Optimization (ACO) are used to provide an efficient estimate of the optimal margin extent in each direction. ACO can provide global optimal solutions in highly nonlinear problems such as margin estimation. The proposed method is demonstrated using cases from stomach lymphoma with daily TomoTherapy megavoltage CT (MVCT) contours. Results: The process consists of two main steps that involve deriving organ motion probability and estimating the optimal margin. For motion probability, we used methods based on rigid registration. The estimation of the optimal margin was carried out using swarm intelligence based on the ACO algorithm. Preliminary results using Dice similarity index are promising. In order to achieve 95% coverage, the ACO ran for 25 nest relocations and converged in about 10 iterations with an optimal Dice metric of 0.88. It is expected that the proposed method will be helpful for guiding future margin estimation protocols. Conclusions: In this work, we have presented a new software tool and an algorithm for estimating organ motion margins from daily images. Our results indicate that the developed tool can provide improved visualization and quantification of daily deformations for estimating isotropic and anisotropic margins for radiotherapy treatment plans in cases where organ motion is an issue such as the presented example of stomach lymphoma, or in cases of lung and prostate cancers.

Dimensional change measurement and stain segmentation in printed fabrics

Dimensional changes and soil release are important functional attributes of fabrics that impact their pricing in the marketplace. This paper presents the underlying principles of a machine vision system that creates a framework for making both measurements in a simultaneous and automatic fashion from digitized fabric images. In addition, the proposed approach is equipped to operate on solid color, as well as printed/patterned fabrics. This system, using image analysis, can (i) measure dimensional changes, which include shrinkage and skew, and (ii) localize stains on the fabric image, which can then be evaluated for soil release. Robust extraction of the above measurements on printed fabric images are made possible using a sequence of customized image registration and background-subtraction techniques. The system was validated using a set of 623 fabric images that involved the detection of 1868 shrinkage dots which were used to make 934 shrinkage measurements, 467 skew measurements, and detect 240 potential stains. The system produced excellent results with a successful shrinkage dot detection rate of 98.9% and an average stain segmentation accuracy of 0.87 using the dice metric.

EM Segmentation of the Distal Femur and Proximal Tibia: A High-Throughput Approach to Anatomic Surface Generation

Fully automated segmentation of computed tomography (CT) images remains a challenge for musculoskeletal researchers. The surfaces generated from image segmentations are valuable for surgical evaluation and planning. Previously, we demonstrated the expectation maximization (EM) algorithm as a semi-automated method of bone segmentation from CT images. In this work, we improve upon the methodology of probability map generation and demonstrate extended applicability of EM-based segmentation to the distal femur and proximal tibia using 72 CT image sets. We also compare the resulting EM segmentations to manual tracings using overlap metrics and time. In the case of the distal femur, the resulting quality metrics had mean values of 0.91 and 0.95 for the Jaccard and Dice metrics, respectively. For the proximal tibia, the Jaccard and Dice metrics were 0.90 and 0.95, respectively. The EM segmentation method was 8 times faster than the average manual segmentation and required less than 4% of the human rater time. Overall, the EM algorithm offers reliable image segmentations with an increased efficiency in comparison to manual segmentation techniques.

Semi-automated cardiac segmentation on cine magnetic resonance images using GVF-Snake deformable models

The segmentation of left ventricular structures is necessary for the evaluation of the ejection fraction (EF) and the myocardial mass (LVM). A semi-automated 2D algorithm using connected filters and a deformable model allowing an accurate endocardial detection was proposed. The epicardial border was deduced using a deformable model restricted inside a region of interest defined from the endocardial border. Papillary muscles were detected using a fuzzy k-means algorithm. The method was applied to the challenge training and validation databases, consisting of 15 subjects each. The evaluation was performed using the tools provided by the challenge. For both datasets, results show a mean Dice metric of 0.89 for endocardial borders (0.92 for epicardial borders). Overall average perpendicular distance was 2.2 mm. Very good correlation was obtained for the EF and LVM parameters. Visual overall rating given by the challenge’s cardiologist was 1.2. Segmentation was robust and performed successfully on both datasets.

Random amplified polymorphic DNA analysis of the phenetic relationships among world-wide accessions of Hydrilla verticillata

The phenetic relationships among forty-four accessions of Hydrilla verticillata from various regions of the world were determined using random amplified polymorphic DNA (RAPD) analysis of bulked genomic samples. Five primers were used producing a total of 85 resolvable, polymorphic bands. The accessions were compared using Gower and Dice metrics, clustered using unweighted pair-group arithmetic average clustering (UPGMA) and consensus algorithms, and factored using principal coordinate analysis. Four major clusters (Asian, Australian, Indonesian, monoecious U.S.) and one minor outlier cluster (Japan/Poland) were identified. The U.S. dioecious accessions formed a group closest to an accession from Bangalore, India, possibly lending credence to historical reports that it was imported from Sri Lanka. The U.S. monoecious plants cluster with an accession from Seoul, Korea. Accessions from Taiwan, Burundi, and Panama join the Asian cluster late. The New Zealand accessions cluster loosely with those from Australia. The use of band intensity in combination with the Gower similarity coefficient generated a cophenetic correlation coefficient (similarity matrix vs. UPGMA matrix) of r = 0.92, superior to that for the corresponding Dice metric ( r = 0.85).