Showing papers by "Juan Manuel Górriz published in 2010"

Computer-aided diagnosis of Alzheimer's disease using support vector machines and classification trees

Abstract: This paper presents a computer-aided diagnosis technique for improving the accuracy of early diagnosis of Alzheimer-type dementia. The proposed methodology is based on the selection of voxels which present Welch's t-test between both classes, normal and Alzheimer images, greater than a given threshold. The mean and standard deviation of intensity values are calculated for selected voxels. They are chosen as feature vectors for two different classifiers: support vector machines with linear kernel and classification trees. The proposed methodology reaches greater than 95% accuracy in the classification task.

Selecting Regions of Interest in SPECT Images Using Wilcoxon Test for the Diagnosis of Alzheimer's Disease

Abstract: This work presents a computer-aided diagnosis technique for improving the accuracy of the diagnosis of the Alzheimer’s disease (AD). Some regions of the SPECT image discriminate more between healthy and AD patients than others, thus, it is important to design an automatic tool for selecting these regions. This work shows the performance of the Mann-Whitney-Wilcoxon U-test, a non-parametric technique which allows to select voxels of interest. Those voxels with higher U values are selected and their intensity values are used as input for a Support Vector Machine classifier with linear kernel. The proposed methodology yields an accuracy greater than 90% in the diagnosis of the AD and outperforms existing techniques including the voxel-as-features approach.

Alzheimer's disease detection in functional images using 2D Gabor wavelet analysis

Abstract: Presented is a Gabor wavelet (GW) based analysis of functional brain images by integrating the 2D GW representation of the images for image classification applied to early diagnosis of Alzheimer's disease. The 2D GW representation of the brain images is processed by means of a principal component analysis (PCA) for feature extraction and support vector machines (SVMs) for image classification. The proposed method yields up to 96% classification accuracy with 100% sensitivity, thus becoming an accurate method for image classification. Comparison between the conventional PCA plus SVM method and the proposed method is also provided. In addition, the proposed method with Gabor wavelets increases the outcomes of other methods based on voxel as features (VAF), PCA, and so on.

Early Alzheimer's disease diagnosis using partial least squares and random forests

Abstract: Currently, the accurate diagnosis of the Alzheimer disease (AD) still remains a challenge in the clinical practice. This paper shows a novel computer aided diagnosis (CAD) system for the early Alzheimer's disease using single photon emission computed tomography (SPECT) images. The proposed system combines a partial least square (PLS) regression model for feature extraction and a random forest predictor. The generalization error of the random forest classifier converges to a limit as the number of trees in the forest increases. PLS feature extraction is found to be more effective for obtaining discriminant information from the data and outperforms principal component analysis (PCA) as a feature extraction technique yielding peak values of sensitivity=100%, specificity= 92.7% and accuracy= 96.9%. Moreover, the proposed CAD system outperformed recently developed AD CAD systems.

Effective diagnosis of alzheimer's disease by means of association rules

Abstract: In this paper we present a novel classification method of SPECT images for the early diagnosis of the Alzheimer's disease (AD) The proposed method is based on Association Rules (ARs) aiming to discover interesting associations between attributes contained in the database The system uses firstly voxel-as-features (VAF) and Activation Estimation (AE) to find tridimensional activated brain regions of interest (ROIs) for each patient These ROIs act as inputs to secondly mining ARs between activated blocks for controls, with a specified minimum support and minimum confidence ARs are mined in supervised mode, using information previously extracted from the most discriminant rules for centering interest in the relevant brain areas, reducing the computational requirement of the system Finally classification process is performed depending on the number of previously mined rules verified by each subject, yielding an up to 95.87% classification accuracy, thus outperforming recent developed methods for AD diagnosis.

Machine learning for very early Alzheimer's Disease diagnosis; a 18 F-FDG and PiB PET comparison

Abstract: This paper shows a machine learning approach based on Principal Component Analysis (PCA) and Support Vector Machine (SVM) to compare the diagnostic accuracy on very early Alzheimer's Disease (AD) patients with 18F FDG and Pittsburg Compound B (PiB) PET imaging. The Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset is used for testing, making use of the longitudinal character. Mild Cognitive Impairment (MCI) individuals that after a two years follow up converted into possible AD where used as very early AD patients. While 18F FDG and PiB have similar diagnostic accuracy in AD, PiB is shown to have higher discriminative power in very early AD with respect to FDG.

Elitist genetic algorithm guided by higher order statistic for blind separation of digital signals

Abstract: A novel method for blind separation of digital signals based on elitist genetic algorithms is presented in this paper. Contrast function, consisting in a weighted sum of high order statistics measures (cumulants of different orders), plays the role of genetic fitness function, and also guide the genetic algorithm by a Gauss-Newton adaptation applied to the genetic population, that reduces the search space and provide faster convergence rate. The use of elitism assures the convergence of the algorithm. Several experiments were conducted on digital signals and mixing models, and the high amount of simulations derived from them provided the best combination of the constant parameters in terms of separation accuracy and convergence rate. In this sense, we also achieve a robust blind source separation method that efficiently adapts to the statistical nature of the mixing signals, within a low population of the genetic algorithm.

Exploratory matrix factorization for PET image analysis

Abstract: Features are extracted from PET images employing exploratory matrix factorization techniques, here non-negative matrix factorization (NMF) Appropriate features are fed into classifiers such as support vector machine or random forest An automatic classification is achieved with high classification rate and only few false negatives.

Sensor fusion adaptive filtering for position monitoring in intense activities

Abstract: Inertial sensors are widely used in body movement monitoring systems Different factors derived from the sensors nature, such as the Angle Random Walk (ARW), and dynamic bias lead to erroneous measurements Moreover, routines including intense exercises are subject to high dynamic accelerations that distort the angle measurement Such negative effects can be reduced through the use of adaptive filtering based on sensor fusion concepts Most existing published works use a Kalman filtering sensor fusion approach Our aim is to perform a comparative study among different adaptive filters Several Least Mean Squares (LMS) and Recursive Least Squares (RLS) filters variations are tested with the purpose of finding the best method leading to a more accurate angle measurement An angle wander compensation and dynamic acceleration bursts filtering method has been developed by the implementing a sensor fusion approach based on LMS and RLS filters.

Improving the convergence rate in affine registration of PET brain images using histogram matching

Abstract: In this work, a procedure to improve the convergence rate for affine registration methods of medical brain images when the images differ greatly from the template is presented. The methodology is based on a histogram matching of the source images with respect to the reference brain template before to proceed with the affine registration. The preprocessed source brain images are spatially normalize to a template using a general affine model with 12 parameters. A sum of squared differences between the source images and the template is considered as objective function and a Gauss-Newton optimization algorithm is used to find the minimum of the cost function. Using histogram equalization as a preprocessing step improves the convergence rate in the affine registration algorithm of brain images as we show in this work using fusion PET/MRI brain images.

Partial least squares for feature extraction of SPECT images

Abstract: Single Photon Emission Computed Tomography (SPECT) images are commonly used by physicians to assist the diagnosis of several diseases such as Alzheimer's disease (AD) The diagnosis process requires the visual evaluation of the image and usually entails time consuming and subjective steps In this context, computer aided diagnosis (CAD) systems are desired This work shows a complete CAD system that uses SPECT images for the automatic diagnosis of AD and combines of support vector machine (SVM) learning with a novel methodology for feature extraction based on the partial least squares (PLS) regression model This methodology avoids the well-known small sample size problem that multivariate approaches suffer and yields peak accuracy rates of 959% The results achieved are compared with the obtained ones by an PCA-based CAD system which is used as baseline

Exploring symmetry to assist alzheimer's disease diagnosis

Abstract: Alzheimer's disease (AD) is a progressive neurodegenerative disorder first affecting memory functions and then gradually affecting all cognitive functions with behavioral impairments and eventually causing death Functional brain imaging as Single-Photon Emission Computed Tomography (SPECT) is commonly used to guide the clinician's diagnosis The essential left-right symmetry of human brains is shown to play a key role in coding and recognition In the present work we explore the implications of this symmetry in AD diagnosis, showing that recognition may be enhanced when considering this latent symmetry.