Catálogo de publicaciones - libros

In this paper we present a new method of 3D non-negative tensor factorization (NTF) that is robust in the presence of noise and has many potential applications, including multi-way blind source separation (BSS), multi-sensory or multi-dimensional data analysis, and sparse image coding. We consider alpha- and beta-divergences as error (cost) functions and derive three different algorithms: (1) multiplicative updating; (2) fixed point alternating least squares (FPALS); (3) alternating interior-point gradient (AIPG) algorithm. We also incorporate these algorithms into multilayer networks. Experimental results confirm the very useful behavior of our multilayer 3D NTF algorithms with multi-start initializations.

In this paper, the design and test results of a QRS complex detector are presented. The detection algorithm is based on the Discrete Wavelet Transform and implements an adaptive weighting scheme of the selected transform coefficients in the time domain. It was tested against a standard MIT-BIH Arrhythmia Database of ECG signals for which sensitivity () of 99.54% and positive predictivity (+ ) of 99.52% was achieved. The designed QRS complex detector is implemented on TI TMS320C6713 DSP for real-time processing of ECGs.

Segmentation for the region of nucleus in the image of uterine cervical cytodiagnosis is known as the most difficult and important part in the automatic cervical cancer recognition system. In this paper, the nucleus region is extracted from an image of uterine cervical cytodiagnosis using the HSI model. The characteristics of the nucleus are extracted from the analysis of morphemetric features, densitometric features, colormetric features, and textural features based on the detected region of nucleus area. The classification criterion of a nucleus is defined according to the standard categories of the Bethesda system. The fuzzy c-means clustering algorithm is employed to the extracted nucleus and the results show that the proposed method is efficient in nucleus recognition and uterine cervical Pap-Smears extraction.

Chest radiographs play an important role in the diagnosis of lung cancer. Detection of pulmonary nodules in chest radiographs forms the basis of early detection. Due to its sparse bone structure and overlapping of the nodule with ribs and clavicles the nodule is hard to detect in conventional chest radiographs. We present a technique based on Independent Component Analysis (ICA) for the suppression of posterior ribs and clavicles which will enhance the visibility of the nodule and aid the radiologist in the diagnosis process.

Hand-based personal identification is a stable and reliable biometrically technique in the field of personal identity recognition. In this paper, both hand shape and palmprint texture features are extracted to facilitate a coarse-to-fine dynamic identification task. The wavelet zero-crossing method is first used to extract hand shape features to guide the fast selection of a small set of similar candidates from the database. Then, a circular Gabor filter, which is robust against brightness, and modified Zernike moments methods are used to extract the features of palmprint. And one-class-one-network (Back-Propagation Neural Network (BPNN) classification structure is employed for final classification. The experimental results show the effectiveness and accuracy of the proposed approach.

The paper presents the automatic system for white blood cell recognition on the basis of the image of bone marrow smear. The paper proposes the complete system solving all problems, beginning from cell extraction using watershed algorithm, generation of different features based on texture, geometry and the statistical description of image intensity, feature selection using Linear Support Vector Machine and final classification by applying Gaussian kernel Support Vector Machine. The results of numerical experiments of recognition of 10 classes of blood cells of patients suffering from leukaemia have shown that the proposed system is sufficiently accurate so as to find practical application in the hospital practice.

Kernel methods are a class of algorithms for pattern analysis with a number of convenient features. This paper proposes extension of the kernel method for biological screening data including chemical compounds. Our investigation of extending kernel aims to combine properties of graphical structure and molecule descriptors. The use of such kernels allows comparison of compounds, not only on graphs but also on important molecular descriptors. Our experimental evaluation of eight different classification problems shows that a proposed special kernel, which takes into account chemical molecule structure and molecule descriptors, statistically improves significantly the classification performance.

In this paper, we proposed a hybrid automated detection system based least square support vector machine (LSSVM) and -NN based weighted pre-processing for diagnosing of macular disease from the pattern electroretinography (PERG) signals. -NN based weighted pre-processing is pre-processing method, which is firstly proposed by us. The proposed system consists of two parts: -NN based weighted pre-processing used to weight the PERG signals and LSSVM classifier used to distinguish between healthy eye and diseased eye (macula diseases). The performance and efficiency of proposed system was conducted using classification accuracy and 10-fold cross validation. The results confirmed that a hybrid automated detection system based on the LSSVM and -NN based weighted pre-processing has potential in detecting macular disease. The stated results show that proposed method could point out the ability of design of a new intelligent assistance diagnosis system.

This paper describes automatic analysis of microscopic mast cell images using network of synchronized oscillators. This network allows for detection of image objects and their boundaries along with evaluation of some geometrical object parameters, like their area and perimeter. Estimation of such mast cells’ parameters is very important in description of both physiological and pathological processes in the human organism. It was also demonstrated, that oscillator networks is able to perform basic morphological operations along with binary image segmentation. Analysis of sample mast cell image was presented and discussed.

DNA analysis by microarrays is a powerful tool that allows replication of the RNA of hundreds of thousands of genes at the same time, generating a large amount of data in multidimensional space that must be analyzed using informatics tools. Various clustering techniques have been applied to analyze the microarrays, but they do not offer a systematic form of analysis. This paper proposes the use of Gorban’s Elastic Neural Net in an iterative way to find patterns of expressed genes. The new method proposed (Iterative Elastic Neural Net, IENN) has been evaluated with up-regulated genes of the Escherichia Coli bacterium and is compared with the Self-Organizing Maps (SOM) technique frequently used in this kind of analysis. The results show that the proposed method finds 86.7% of the up-regulated genes, compared to 65.2% of genes found by the SOM. A comparative analysis of Receiver Operating Characteristic (ROC) with SOM shows that the proposed method is 11.5% more effective.