Attention Deficit Hyperactivity Disorder Diagnosis using non-linear univariate and multivariate EEG measurements: a preliminary study.

Attention Deficit Hyperactivity Disorder (ADHD) is a common neuro-developmental disorder of childhood. In this study we propose two classification algorithms for discriminating ADHD children from normal children using their resting state Electroencephalography (EEG) signals. One algorithm is based on the univariate features extracted from individual EEG recording channels and the other is based on the multivariate features extracted from brain lobes. We focused on entropy measures as non-linear univariate and multivariate features. Average power, Theta/Beta Ratio (TBR), Shannon Entropy (ShanEn), Sample Entropy (SampEn), Dispersion Entropy (DispEn) and Multiscale SampEn (MSE) were extracted as linear and non-linear univariate features. Besides, multivariate SampEn (mvSE) and multivariate MSE (mvMSE) were extracted as non-linear multivariate features. Classification was followed by three classifiers: Support Vector Machines (SVM) with different kernels, k-Nearest Neighbor (kNN) and Probabilistic Neural Network (PNN). Complexity analysis of multi-channel EEG data was performed using mvMSE approach. Entropy mapping as a useful tool was used to visually track changes of entropies in various brain regions. Based on achieved results, ADHD children have higher brain activity and TBR compared to normal children, while their neural system is more regular. Besides, ADHD children have reduced dynamical complexity of neural system. Finally, the accuracy of 99.58% was achieved in classification based on a combination of non-linear univariate features by Radial Basis Function (RBF) SVM. For classification based on brain regions using multivariate features, 90.63% accuracy was achieved by PNN.

A novel multi-scale Hessian based spot enhancement filter for two dimensional gel electrophoresis images.

Two dimensional gel electrophoresis (2DGE) is a useful method for studying proteins in a wide variety of applications including identifying post-translation modification (PTM), biomarker discovery, and protein purification. Computerized segmentation and detection of the proteins are the two main processes that are carried out on the scanned image of the gel. Due to the complexities of 2DGE images and the presence of artifacts, the segmentation and detection of protein spots in these images are non-trivial, and involve supervised and time consuming processes. This paper introduces a new spot filter for enhancing, and separating the closely overlapping spots of protein in 2DGE images based on the multi-scale eigenvalue analysis of the image Hessian. Using a Gaussian spot model, we have derived closed form equations to compute the eigen components of the image Hessian of two overlapping spots in a multi-scale fashion. Based on this analysis, we have proposed a novel filter that suppresses the overlapping area and results in a better spot separation. The performance of the proposed filter has been evaluated on the synthetic and real 2DGE images. The comparison with three conventional techniques and a commercial software package reveals the superiority and effectiveness of the proposed filter.