Catálogo de publicaciones - libros

Cardiac dynamics suppression is a main issue for visual improvement and computation of tissue mechanical properties in IntraVascular UltraSound (IVUS). Although in recent times several motion compensation techniques have arisen, there is a lack of objective evaluation of motion reduction in pullbacks. We consider that the assessment protocol deserves special attention for the sake of a clinical applicability as reliable as possible. Our work focuses on defining a quality measure and a validation protocol assessing IVUS motion compensation. On the grounds of continuum mechanics laws we introduce a novel score measuring motion reduction in sequences. Synthetic experiments validate the proposed score as measure of motion parameters accuracy; while results in pullbacks show its reliability in clinical cases.

An image analysis system (IAS) was developed for the quantitative assessment of estrogen receptor’s (ER) positive status from breast tissue microscopy images. Twenty-four cases of breast cancer biopsies, immunohistochemically (IHC) stained for ER, were microscopically assessed by a histopathologist, following a clinical routine scoring protocol. Digitized microscopy views of the specimens were used in the IAS’s design. IAS comprised a/image segmentation, for nuclei determination, b/extraction of textural features, by processing of nuclei-images utilizing the Laws and Gabor filters and by calculating textural features from the processed nuclei-images, and c/PNN and SVM classifiers design, for discriminating positively stained nuclei. The proportion of the latter in each case’s images was compared against the physician’s score. Using Spearman’s rank correlation, high correlation was found between the histopathogist’s and IAS’s scores (rho=0.89, p<0.001) and 22/24 cases were correctly characterised, indicating IAS’s reliability in the quantitative evaluation of ER as additional assistance to physician’s assessment.

This paper is a joint effort between five institutions that introduces several novel similarity measures and combines them to carry out a multimodal segmentation evaluation. The new similarity measures proposed are based on the location and the intensity values of the misclassified voxels as well as on the connectivity and the boundaries of the segmented data. We show experimentally that the combination of these measures improves the quality of the evaluation, increasing the significance between different methods both visually and numerically and providing better understanding about their difference. The study shown here has been carried out using four different segmentation methods applied to a MRI simulated dataset of the brain.

Lung segmentation is a necessary first step to computer analysis in lung CT. It is crucial to develop automated segmentation algorithms capable of dealing with the amount of data produced in thin slice multidetector CT and also to produce accurate border delineation in cases of high density pathologies affecting the lung border. In this study an automated method for lung segmentation of thin slice CT data is proposed. The method exploits the advantage of a wavelet preprocessing step in combination with the minimum error thresholding technique applied on volume histogram. Performance averaged over left and right lung volumes is in terms of: lung volume overlap 0.983 ± 0.008, mean distance 0.770 ± 0.251 mm, rms distance 0.520 ± 0.008 mm and maximum distance differentiation 3.327 ± 1.637 mm. Results demonstrate an accurate method that could be used as a first step in computer lung analysis in CT.

The quantitative description of the cardiac motion is an important task for the assessment of the viability in the heart wall. Abnormalities in heart motion can eventually lead to life threatening cardiac injuries therefore measurements of dynamic heart functions are of great clinical importance. In this work we present a method for estimating heart motion from 4D (3D+time) echocardiographic images. Our approach involves non-linear 4D anisotropic diffusion filtering of the data, non-rigid registration of time sequence, noise removal by time averaging, shape and texture based segmentation and finally the reconstruction of the dynamic 3D model of the heart.

In histomorphometrical investigations of bone tissue modeling around screw-shaped implants, the manual measurements of bone area and bone-implant contact length around the implant are time consuming and subjective. In this paper we propose an automatic image analysis method for such measurements. We evaluate different discriminant analysis methods and compare the automatic method with the manual one. The results show that the principal difference between the two methods occurs in length estimation, whereas the area measurement does not differ significantly. A major factor behind the dissimilarities in the results is believed to be misclassification of staining artifacts by the automatic method.

The image segmentation plays an important role in medical image processing. Many segmentation algorithms exist. Most of them produce raster data which is not suitable for 3D geometrical modeling of human tissues. In this paper, a vector segmentation algorithm based on a 3D Delaunay triangulation is proposed. Tetrahedral mesh is used to divide a volumetric CT/MR data into non-overlapping regions whose characteristics are similar. Novel methods for improving quality of the mesh and its adaptation to the image structure are also presented.

Penalized-Likelihood (PL) image reconstruction methods produce better quality images than analytical methods. However, these methods produce images with non-uniform resolution properties. A number of prior functions have been used to recover for this reconstructed resolution non-uniformity. Quadratic Priors (QPs) are preferred due to their simplicity and their resolution characteristics have been studied extensively. Images reconstructed using QPs still exhibit non-uniform resolution properties.Here, we propose median priors (MPs) in place of QPs and evaluate their resolution characteristics.Although, they produce images with non-uniform reconstructed resolution, their recovered resolution is better than the QPs.We have also implemented MPs in a modified penalty frame work, proposed for QPs, and have shown that they produce images with almost uniform resolution. Due to their automatic edge preservation and better quantitative properties, MPs might be preferred over QPs for penalize-likelihood image reconstruction.

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.

Intestinal contractions are one of the main features for analyzing intestinal motility and detecting different gastrointestinal pathologies. In this paper we propose Eigenmotion-based Contraction Detection (ECD), a novel approach for automatic annotation of intestinal contractions of video capsule endoscopy. Our approach extracts the main motion information of a set of contraction sequences in form of eigenmotions using Principal Component Analysis. Then, it uses a selection of them to represent the high dimension motion data. Finally, this contraction characterization is used to classify the contraction sequences by means of machine learning techniques. The experimental results show that motion information is useful in the contraction detection. Moreover, the proposed automatic method is essential to speed up the costly examination of the video capsule endoscopy.