Antonio Bravo

An unsupervised clustering framework for automatic segmentation of left ventricle cavity in human heart angiograms

Cardiac function is routinely assessed from X-rays angiograms acquired at the cardiac catheterization rooms. Currently, the evaluation of cardiac function involves the global measurement of volumes and ejection fraction (EF). This evaluation requires the segmentation of the left ventricle (LV) contour. Several automatic segmentation methods have been reported, however, they are not yet fully validated and accepted in the clinical work. This paper reports on an automatic segmentation method for the ventricular cavity in mono-plane and bi-plane ventriculographic image sequences. The first step is the preprocessing, where a linear regression model is applied to exploit the functional relationship between the original input image and its smoothed version. A two stage clustering algorithm is used for segmenting the left ventricle cavity. First, an approximate initial segmentation is achieved by using a simple linkage region growing algorithm on the preprocessed version of the input image. The second stage is based on a region growing method by multiple linkage. This second stage is intended for refining the initial approximate segmentation. A validation is performed by comparing the estimated contours with respect to contours traced manually by several cardiologists. The average positioning error considering 15 mono-plane and 3 bi-plane angiographic sequences is 0.72mm at end-diastole (ED) and 0.91mm at end-systole (ES). The average contour error is 6.67% at ED and 12.44% at ES. The average area error is 8.58% at ED and 3.32% at ES. The left ventricle volume and the ejection fraction are estimated from manual contours and from the estimated contours showing an excellent correlation: 0.999 for ED volume, 0.998 for ES volume, and 0.952 for EF.

Myocardial border detection from ventriculograms using support vector machines and real-coded genetic algorithms

In this research a two step method for left ventricle segmentation based on landmark detection and evolutionary snakes is reported. The proposed approach is applied to human heart angiograms. Several anatomical landmarks located on the left ventricle are obtained using support vector machines. The training stage is performed based on a set of windows of size 31 x 31 including landmarks and non-landmarks pixel patterns. The support vector machines use a radial basis function kernel and the structural risk minimization principle as the inference rule. During the training stage, no false positives are obtained and during the detection stage a 97.94% of recognition is attained. The estimated landmark location is used for constructing an approximate myocardial border. This contour is a deformable model that is optimized using a real-coded genetic algorithm. A validation is performed by comparing the estimated contours with respect to contours manually traced by two cardiologists. From this validation stage the maximum of the average contour error considering 6 angiographic sequences (a total of 178 images) is 4.93%.

A clustering based approach for automatic image segmentation: an application to biplane ventriculograms

This paper reports on an automatic method for ventricular cavity segmentation in angiographic images. The first step of the method consists in applying a linear regression model that exploits the functional relationship between the original input image and a smoothed version. This intermediate result is used as input to a clustering algorithm, which is based on a region growing technique. The clustering algorithm is a two stage process. In the first stage an initial segmentation is achieved using as input the result of the linear regression and the smoothed version of the input image. The second stage is intended for refining the initial segmentation based on feature vectors including the area, the gray-level average and the centroid of each candidate region. The segmentation method is conceptually simple and provides an accurate contour detection for the left ventricle cavity.

Edge detection in ventriculograms using support vector machine classifiers and deformable models

In this paper a left ventricle (LV) contour detection method is described. The method works from an approximate contour defined by anatomical landmarks extracted using Support Vector Machine (SVM) classifiers. The LV contour approximation is used as an initialization step for the deformable model algorithm. An optimization method based on a gradient descend algorithm is used to obtain the optimal contour that provides a minimum energy value. Both classifier and edge detection method performances have been validated. The error is determined as the difference between the shape estimated by the algorithm and the shape traced by an expert.

An approach to coronary vessels detection in X-ray rotational angiography

An unsupervised clustering framework for automatic detection of coronary vessels in bidimensional (2D) X-ray rotational angiography is reported. The proposed approach consists of three consecutive steps: 1) vessel enhancement; 2) initial segmentation based on a simple linkage region growing algorithm; 3) optimization of the initial segmentation using a multiple linkage region growing method. Results obtained after applying this method to monoplane rotational X-ray image sequences are presented.

Estimation of the deformation field for the left ventricle walls in 4-d multislice computerized tomography

This paper describes a method for estimating the deformation field of the Left Ventricle (LV) walls from a 4–D Multi Slice Computerized Tomography (MSCT) database. The approach is composed of two stages: in the first, a 2–D non–rigid correspondence algorithm matches a set of contours on the LV at consecutive time instants. In the second, a 3–D curvature–based correspondence algorithm is used to optimize the initial approximate correspondence. The dense displacement field is obtained based on the optimized correspondence. Parameters like LV volume, radial contraction and torsion are estimated. The algorithm is validated on synthetic objects and tested using a 4–D MSCT database. Results are promising as the error of the displacement vectors is 2.69 ± 1.38 mm using synthetic objects and, when tested in real data, local parameters extracted agree with values obtained using tagged magnetic resonance imaging.

Inferring the left ventricle dynamical behavior using a free-form deformations model

A computational 4D (3D+time) model for simulating the dynamical shape of the left ventricle (LV) based on free-form deformations (FFD) techniques is described. The simulation model is useful as a teaching tool for understanding the normal left ventricle motion. The model is also useful for initializing 3D segmentation algorithms and for understanding the relation between pathologies and variation of parameters defining the ventricular function. Validation of this computational model is performed by synthesizing 4D sequences of the left ventricle, comprising the interval going from end-systole to end-diastole. From the resulting 4D shapes, several mechanical parameters such as the left ventricle volume, the radial contraction and torsion are calculated and compared with results of works previously reported based in MR-tagging images. A comparison is also performed with respect to mechanical parameters extracted from the additional time instants in the same multislice computerized tomography (MSCT) database used for extracting the LV wall surfaces required for initialization. First results show a good match between parameters compared.

Computational assessment of stomach tumor volume from multi-slice computerized tomography images in presence of type 2 cancer

Background: The multi-slice computerized tomography (MSCT) is a medical imaging modality that has been used to determine the size and location of the stomach cancer. Additionally, MSCT is considered the best modality for the staging of gastric cancer. One way to assess the type 2 cancer of stomach is by detecting the pathological structure with an image segmentation approach. The tumor segmentation of MSCT gastric cancer images enables the diagnosis of the disease condition, for a given patient, without using an invasive method as surgical intervention. Methods: This approach consists of three stages. The initial stage, an image enhancement, consists of a method for correcting non homogeneities present in the background of MSCT images. Then, a segmentation stage using a clustering method allows to obtain the adenocarcinoma morphology. In the third stage, the pathology region is reconstructed and then visualized with a three-dimensional (3-D) computer graphics procedure based on marching cubes algorithm. In order to validate the segmentations, the Dice score is used as a metric function useful for comparing the segmentations obtained using the proposed method with respect to ground truth volumes traced by a clinician. Results: A total of 8 datasets available for patients diagnosed, from the cancer data collection of the project, Cancer Genome Atlas Stomach Adenocarcinoma (TCGASTAD) is considered in this research. The volume of the type 2 stomach tumor is estimated from the 3-D shape computationally segmented from the each dataset. These 3-D shapes are computationally reconstructed and then used to assess the morphopathology macroscopic features of this cancer. Conclusions: The segmentations obtained are useful for assessing qualitatively and quantitatively the stomach type 2 cancer. In addition, this type of segmentation allows the development of computational models that allow the planning of virtual surgical processes related to type 2 cancer.

Similarity enhancement for automatic segmentation of cardiac structures in computed tomography volumes

The aim of this research is proposing a 3-D similarity enhancement technique useful for improving the segmentation of cardiac structures in Multi-Slice Computerized Tomography (MSCT) volumes. The similarity enhancement is obtained by subtracting the intensity of the current voxel and the gray levels of their adjacent voxels in two volumes resulting after preprocessing. Such volumes are: a. — a volume obtained after applying a Gaussian distribution and a morphological top-hat filter to the input and b. — a smoothed volume generated by processing the input with an average filter. Then, the similarity volume is used as input to a region growing algorithm. This algorithm is applied to extract the shape of cardiac structures, such as left and right ventricles, in MSCT volumes. Qualitative and quantitative results show the good performance of the proposed approach for discrimination of cardiac cavities.

Three-Dimensional Segmentation of Ventricular Heart Chambers from Multi-Slice Computerized Tomography: An Hybrid Approach

This research is focused on segmentation of the heart ventricles from volumes of Multi Slice Computerized Tomography (MSCT) image sequences. The segmentation is performed in three–dimensional (3–D) space aiming at recovering the topological features of cavities. The enhancement scheme based on mathematical morphology operators and the hybrid–linkage region growing technique are integrated into the segmentation approach. Several clinical MSCT four dimensional (3–D + t) volumes of the human heart are used to test the proposed segmentation approach. For validating the results, a comparison between the shapes obtained using the segmentation method and the ground truth shapes manually traced by a cardiologist is performed. Results obtained on 3–D real data show the capabilities of the approach for extracting the ventricular cavities with the necessary segmentation accuracy.