Catálogo de publicaciones - libros

This work analyzes the application of the semivariogram function to the characterization of breast tissue as malignant or benign in mammographic images. The method characterization is based on a process that selects, using stepwise technique, from all computed semivariance which best discriminate between the benign and malignant tissues. Then, a multilayer perceptron neural network is used to evaluate the ability of these features to predict the classification for each tissue sample. To verify this application we also describe tests that were carried out using a set of 117 tissues samples, 67 benign and 50 malignant. The result analysis has given a sensitivity of 92.8%, a specificity of 83.3% and an accuracy above 88.0%, which means encouraging results. The preliminary results of this approach are very promising in characterizing breast tissue.

A major problem that researchers face when working with digital medical images is that not all the information about the image is stored with it. Image analysis is used to extract useful information from the images. A clear problem when analyzing digital mammograms is that there is not a direct method to interpret pixel values. In this paper a method is presented that, performs the estimation of the amount of glandular tissue in digital mammograms. This contribution also shows how this knowledge about the pixels allows performing a segmentation of the breast region in its constituent parts; and if present, microcalcifications may be detected.

In this paper focuses on extracting and analyzing spectral features from Trans-Rectal Ultra-Sound (TRUS) images for prostate tissue characterization. The information of the images’ frequency domain features and spatial domain features are used to achieve an accurate Region of Interest (ROI) identification. In particular, each image is divided into ROIs by the use of Gabor filters, a crucial stage, where the image is segmented according to the frequency response of the image pixels. Further, pixels with a similar response to the same filter are assigned to the same region to form a ROI. The radiologist’s experience is also integrated into the algorithm to identify the highly suspected ROIs. Next, for each ROI, different spectral feature sets are constructed. One set includes the power spectrum wedge and ring energies. The other sets are constructed using geometrical features extracted from the Power Spectrum Density (PSD). In particular, the estimated PSD in these sets is divided into two segments. Polynomial interpolation is used for the first segment and the obtained polynomial coefficients are used as features. The second segment is approximated by a straight line and the slope, the Y intercept as well as the first maximum reached by the PSD are considered as features. A classifier-based feature selection algorithm using CLONALG, a recently proposed optimization technique developed on the basis of clonal selection of the Artificial Immune System (AIS), is adopted and used to select an optimal subset from the above extracted features. Using different PSD estimation techniques, the obtained accuracy ranges from 72.2% to 93.75% using a Support Vector Machine classifier.

Follow-up studies of human organs with certain abnormalities are necessary to monitor the effect of therapy and the extent of recovery. But, current approaches are limited to subjective estimation by physicians. In this paper, we present a comparative visualization and objective analysis system for long-term follow-up studies using cardiac MR images. This system uses dynamic cardiac models in order to effectively analyze huge cardiac datasets acquired at different times and applies the data hierarchy that is proper for effectively managing the multi-dimensional cardiac images, dynamic models and analyzed parameters for each patient. The proposed system provides various quantitative analysis functions for global and local diagnostic parameters of ventricles based on the physical properties of the heart.

The aim of this paper is to examine a multiclassifier approach to the classification of the lung nodules in X-ray chest radiographs. The approach investigated here is based on an image region-based classification whose output is the information of the presence or absence of a nodule in an image region. The classification was made, essentially, in two steps: firstly, a set of rotation invariant features was extracted from the responses of a multi-scale and multi-orientation filter bank; secondly, different classifiers (multi-layer perceptrons) are designed using different features sets and trained in different data. These classifiers are further combined in order to improve the classification performance. The obtained results are promising and can be used for reducing the false-positives nodules detected in a computer-aided diagnosis system.

We present a fully automated method for extracting the lung region from volumetric X-ray CT images based on material decomposition. By modeling the human thorax as a composition of different materials , the proposed method follows a threshold-based, hierarchical voxel classification strategy. The segmentation procedure involves the automatic computation of threshold values and consists on three main steps: patient segmentation and decomposition, large airways extraction and lung parenchyma decomposition, and lung region of interest segmentation. Experimental results were performed on thoracic CT images acquired from 30 patients. The method provides a reproducible set of thresholds for accurate extraction of the lung parenchyma, needed for computer aided diagnosis systems.

A collection of 2200 microscopic sections was recently recovered at the Netherlands Ophthalmic Research Institute and the Department of Anatomy and Embryology of the Academic Medical Centre in Amsterdam. The sections were created thirty years ago and constitute the largest and most detailed study of human orbital anatomy to date. In order to preserve the collection, it was digitised. This paper documents a practical approach to the automatic reconstruction of a 3-D representation of the original objects from the digitised sections. To illustrate the results of our approach, we show a multi-planar reconstruction and a 3-D direct volume rendering of a reconstructed foetal head.

This paper presents a computer aided diagnosis system for skin lesions. Diverse parameters or features extracted from fluorescence images are evaluated for cancer diagnosis. The selection of parameters has a significant effect on the cost and accuracy of an automated classifier. The genetic algorithm (GA) performs parameters selection using the classifier of the K-nearest neighbours (KNN). We evaluate the classification performance of each subset of parameters selected by the genetic algorithm. This classification approach is modular and enables easy inclusion and exclusion of parameters. This facilitates the evaluation of their significance related to the skin cancer diagnosis. We have implemented this parameter evaluation scheme adopting a strategy that automatically optimizes the K-nearest neighbours classifier and indicates which features are more relevant for the diagnosis problem.

The paper is an extension of previous work on spatial-varying Gaussian mixture and Markov random field (SVGM-MRF) from 2D to 3D to segment the MR brain volume with the presence of noise and inhomogeneity. The reason for this extension is that MR brain data are naturally three dimensional, and the information from the additional dimension provides a more accurate conditional probability representation. To reduce large computation time and memory requirements for 3D implementation, a method of using only the local window information to perform the necessary parameter estimations and to achieve the tissue labeling is proposed. The experiments on fifteen brain volumes with various noise and inhomogeneity levels and comparisons with other three well-known 2D methods are provided. The new method outperforms all three 2D methods for high noise and inhomogeneity data which is a very common occurrence in MR applications.

The lack of consistent ordering of components resulted from independent component analysis poses a significant obstacle to the pervasive application of this method on fMRI data analysis. Based on the temporal correlation of physiological noise components of fMRI data and that of cerebrospinal fluid data, the ordering of independent spatial components is ranked using canonical correlation analysis. The proposed method can robustly identify the task-related spatial component without any prior information about the functional activation paradigm. The experimental results of analyzing the real fMRI data show the reliability of the presented method.