Inas A. Yassine

A novel technique for cardiac arrhythmia classification using spectral correlation and support vector machines

A method for arrhythmia classification based on spectral correlation is proposed.Statistical features for the spectral correlation coefficients were calculated.Features were subjected to principal component analysis and fisher score.Raw spectral correlation data, PCA data and FS data were classified using SVM.The best performance is achieved using raw spectral correlation data. Cardiac disorders are one of the main causes leading to death. Therefore, they require continuous and efficient detection techniques. ECG is one of the main tools to diagnose cardiovascular disorders such as arrhythmias. Computer aided diagnosis (CAD) systems play a very important role in early detection and diagnosis of cardiac arrhythmias. In this work, we propose a CAD system for classifying five beat types including: normal (N), Premature Ventricular Contraction (PVC), Premature Atrial Contraction (APC), Left Bundle Branch Block (LBBB) and Right Bundle Branch Block (RBBB). The proposed system is based on cyclostationary signal analysis approach, which explores hidden periodicities in the signal of interest and thus it is able to detect hidden features. In order to study the cyclostationarity properties of the signal, we utilized the spectral correlation as a nonlinear statistical transformation inspecting the periodicity of the correlation. Three experiments were investigated in our study; raw spectral correlation data were used in the first experiment while the other two experiments utilized statistical features for the raw spectral data followed by principal component analysis (PCA) and fisher score for feature space reduction purposes respectively. As for the classification task, support vector machine (SVM) with linear kernel was employed for all experiments. The experimental results showed that the approach that uses the raw spectral correlation data is superior compared to several state of the art methods. This approach achieved sensitivity, specificity, accuracy, positive predictive value (PPV) and negative predictive value (NPV) of 99.20%, 99.70%, 98.60%, 99.90% and 97.60% respectively.

Convolutional neural networks for deep feature learning in retinal vessel segmentation

Analysis of retinal vessels in fundus images provides a valuable tool for characterizing many retinal and systemic diseases. Accurate automatic segmentation of these vessels is usually required as an essential analysis step. In this work, we propose a new formulation of deep Convolutional Neural Networks that allows simple and accurate segmentation of the retinal vessels. A major modification in this work is to reduce the intra-class variance by formulating the problem as a Three-class problem that differentiates: large vessels, small vessels, and background areas. In addition, different sizes of the convolutional kernels have been studied and it was found that a combination of kernels with different sizes achieve the best sensitivity and specificity. The proposed method was tested using DRIVE dataset and it showed superior performance compared to several other state of the art methods. The segmentation sensitivity, specificity and accuracy were found to be 83.97%, 95.62% and 94.56% respectively.

Robust multi-component modeling of diffusion tensor magnetic resonance imaging data

In conventional diffusion tensor imaging (DTI) based on magnetic resonance data, each voxel is assumed to contain a single component having diffusion properties that can be fully represented by a single tensor. In spite of its apparent lack of generality, this assumption has been widely used in clinical and research purpose. This resulted in situations where correct interpretation of data was hampered by mixing of components and/or tractography. Even though this assumption can be valid in some cases, the general case involves mixing of components resulting in significant deviation from the single tensor model. Hence, a strategy that allows the decomposition of data based on a mixture model has the potential of enhancing the diagnostic value of DTI. This work aims at developing a stable solution for the most general problem of multi-component modeling of diffusion tensor imaging data. This model does not include any assumptions about the nature or volume ratio of any of the components and utilizes a projection pursuit based strategy whereby a combination of exhaustive search and least-squares estimation is used to estimate 1D projections of the solution. Then, such solutions are combined to compute the multidimensional components in a fast and robust manner. The new method is demonstrated by both computer simulations and real diffusion-weighted data. The preliminary results indicate the success of the new method and its potential to enhance the interpretation of DTI data sets.

Novel features for microcalcification detection in digital mammogram images based on wavelet and statistical analysis

Computer Aided Diagnosis (CAD) systems play an important role in early detection of breast cancer. In this study, we propose a CAD system based on a novel feature set for detection of microcalcifications. The new features are inspired from several statistical observations for some classical features such as higher order statistical (HOS) features, Discrete Wavelet Transform (DWT) and Wavelet Packet Decomposition (WPD) based features. Our study employs DWT for preprocessing and Students t-test for evaluation and reduction of the features. Support vector machines (SVM) with linear and RBF kernels was used. The proposed system achieved 98.43%, 96.74% sensitivity, 93.34%, 94.87% specificity and 95.80%, 95.78% accuracy using RBF kernel for MIAS and DDSM databases respectively.

Novel Multi-Tensor Estimation for High-Resolution Diffusion Tensor Magnetic Resonance Imaging

The estimation of diffusion tensors in diffusion tensor imaging (DTI) is based on the assumption that each voxel is homogeneous and can be represented by a single tensor. As a result, estimation errors arise particularly in voxels with partial voluming of white matter or gray matter with cerebrospinal fluid (CSF) and voxels where fibers cross. Several authors have explored the possibility of solving for multiple tensors. Several authors analyzed the problem from the point of view of the number of unknowns and concluded that the solution is rather difficult due to the large number of unknowns and the nonlinearity of the equations. However, their approach was only helpful in eliminating some of the sources of artifacts in the DTI data but offered only a qualitative description of the model components. We implemented three strategies gradient, differentiation and exhaustive algorithms to solve and compare between their estimations at various conditions of signal to noise ratios (SNR)

Low quality dermal image classification using transfer learning

In this study, we investigate three class skin lesion classification problem of a low quality and small size dataset using transfer learning using AlexNet deep Convolutional Neural Network (CNN). Our approach involves modifying the pre-trained AlexNet model; through replacing the decision layer to be compatible with our three class problem. In addition, we propose adding two dropout layers to overcome the over fitting problem. The fine tuning process of the complete network, based on stochastic gradient descent, is performed using skin lesion dataset. Furthermore, we investigated augmenting the original dataset through three flipping directions and sixteen rotation angles processes using a new methodology. The proposed algorithm has been compared with a hand crafted features, based on Local Binary Pattern (LBP) representation followed by Support Vector Machine (SVM) classifier. Increasing the dataset size has dramatically boosted the performance of classifiers achieving accuracy of 98.67% for the modified AlexNet compared to 96.8% using the LBP based system.

Spectral correlation analysis for microcalcification detection in digital mammogram images

Breast cancer has been threatening lives of women around the world. Thus, Computer Aided Diagnosis (CAD) systems play an important role in early detection of breast cancer. In this study, we propose a CAD system based on cyclostationary signal analysis for microcalcifications detection. Spectral correlation is estimated for regions of interests (ROIs) after conversion to 1D vector. The proposed algorithm utilizes simple statistical features calculation for the raw spectral data followed by student-t test for evaluation and reduction of the generated set of features. Support Vector Machines (SVM) with linear kernel was employed for the classification task. The experimental results showed that the proposed approach is superior compared to several state of the art methods. This approach was tested using digital database for screening mammogram (DDSM) and it achieved sensitivity, specificity and accuracy of 95.88%, 93.10% and 94.44% respectively.

Novel Bayesian classifier discriminant function optimization strategies for arrhythmia classification

Cardiac arrhythmia is considered to be one of the most critical addressed problems leading to death. Thus, Computer Aided Diagnosis (CAD) systems are essential for early arrhythmia detection and diagnosis. In this paper, we propose a classification system for arrhythmia diagnosis based on Bayesian classifier. The system employs one-versus-one approach, used in the classification methodology of several multi-class classifiers such as the support Vector Machine (SVM). The proposed idea is mainly based on introducing new algorithms for optimizing the classifiers parameters in order to improve the overall classification system performance, using the Space Search (SS) and the One-versus-One Error Minimization (OOEM) approaches. The SS approach boosted system accuracy over the conventional Bayes (CB) by 1.14%, 2.5% and 3.33% for 3, 5 and 6-classes arrhythmia problems respectively while OOEM showed less superiority than SS as it boosted accuracy by 0.7% and 2.44% for the 5 and 6-classes problems respectively and attained same accuracy achieved by CB for the 3-class problem. The learning and testing times were calculated for both approaches. The results show that the SS based system offers the best possible accuracy, however it has the longest learning time.

Cognitive functions, electroencephalographic and diffusion tensor imaging changes in children with active idiopathic epilepsy

Abstract Introduction Neurocognitive impairment represents one of the most common comorbidities occurring in children with idiopathic epilepsy. Diagnosis of the idiopathic form of epilepsy requires the absence of any macrostructural abnormality in the conventional MRI. Though changes can be seen at the microstructural level imaged using advanced techniques such as the Diffusion Tensor Imaging (DTI). Aim of the work The aim of this work is to study the correlation between the microstructural white matter DTI findings, the electroencephalographic changes and the cognitive dysfunction in children with active idiopathic epilepsy. Methods A comparative cross-sectional study, included 60 children with epilepsy based on the Stanford–Binet 5th Edition Scores was conducted. Patients were equally assigned to normal cognitive function or cognitive dysfunction groups. The history of the epileptic condition was gathered via personal interviews. All patients underwent brain Electroencephalography (EEG) and DTI, which was analyzed using FSL. Results The Fractional Anisotropy (FA) was significantly higher whereas the Mean Diffusivity (MD) was significantly lower in the normal cognitive function group than in the cognitive dysfunction group. This altered microstructure was related to the degree of the cognitive performance of the studied children with epilepsy. The microstructural alterations of the neural fibers in children with epilepsy and cognitive dysfunction were significantly related to the younger age of onset of epilepsy, the poor control of the clinical seizures, and the use of multiple antiepileptic medications. Conclusion Children with epilepsy and normal cognitive functions differ in white matter integrity, measured using DTI, compared with children with cognitive dysfunction. These changes have important cognitive consequences.INTRODUCTION Neurocognitive impairment represents one of the most common comorbidities occurring in children with idiopathic epilepsy. Diagnosis of the idiopathic form of epilepsy requires the absence of any macrostructural abnormality in the conventional MRI. Though changes can be seen at the microstructural level imaged using advanced techniques such as the Diffusion Tensor Imaging (DTI). AIM OF THE WORK The aim of this work is to study the correlation between the microstructural white matter DTI findings, the electroencephalographic changes and the cognitive dysfunction in children with active idiopathic epilepsy. METHODS A comparative cross-sectional study, included 60 children with epilepsy based on the Stanford-Binet 5th Edition Scores was conducted. Patients were equally assigned to normal cognitive function or cognitive dysfunction groups. The history of the epileptic condition was gathered via personal interviews. All patients underwent brain Electroencephalography (EEG) and DTI, which was analyzed using FSL. RESULTS The Fractional Anisotropy (FA) was significantly higher whereas the Mean Diffusivity (MD) was significantly lower in the normal cognitive function group than in the cognitive dysfunction group. This altered microstructure was related to the degree of the cognitive performance of the studied children with epilepsy. The microstructural alterations of the neural fibers in children with epilepsy and cognitive dysfunction were significantly related to the younger age of onset of epilepsy, the poor control of the clinical seizures, and the use of multiple antiepileptic medications. CONCLUSION Children with epilepsy and normal cognitive functions differ in white matter integrity, measured using DTI, compared with children with cognitive dysfunction. These changes have important cognitive consequences.

Schizophrenic patient identification using graph-theoretic features of resting-state fMRI data

Abstract Resting-state functional magnetic resonance imaging (Rs-fMRI) is a promising imaging modality to study the changes of functional brain networks in schizophrenic patients. Several representations have been proposed to capture the essential features of these networks. In particular, graph-theoretic representations can be effectively used to discriminate healthy subjects from schizophrenic patients. In this paper, we propose a machine-learning system based on a graph-theoretic approach to investigate and differentiate the brain network alterations. The fMRI data samples are first preprocessed to reduce noise and normalize the images. The automated anatomical labeling (AAL) atlas is then used to parcellate the brain into 90 regions and construct a region connectivity matrix. A weighted undirected graph is hence constructed and graph measures are computed for each subject. These graph measures include betweenness centrality, characteristic path length, degree, clustering coefficient, local efficiency, global efficiency, participation coefficient and small-worldness. After that, feature selection algorithms are used to choose the most discriminant features. Finally, a SVM classifier is trained and tested on discriminant graph features. Experiments were performed on a large Rs-fMRI dataset formed of 70 schizophrenic patients and 70 healthy subjects. The performance was evaluated using nested-loop 10-fold cross-validation. The best detection results were found using the feature selection methods of Welchs t-test (82.85%), l0-norm (91.43%), and feature selection via concave minimization (FSV) (95.00%). Our results outperform those of recent state-of-the-art graph-theoretic methods.