Full Research Paper
Saeed Ghodsi; Hoda Mohammadzade; Hamid Aghajan
Volume 13, Issue 3 , October 2019, Pages 189-207
Abstract
Different perceptual, cognitive and emotional situations results in a kind of information flow in the brain by means of coordinated neuronal oscillations. Analysing these oscillations, especially synchronizations of different brain regions, can illustrate the brains response in the aforementioned situations. ...
Read More
Different perceptual, cognitive and emotional situations results in a kind of information flow in the brain by means of coordinated neuronal oscillations. Analysing these oscillations, especially synchronizations of different brain regions, can illustrate the brains response in the aforementioned situations. In the literature, connectivity between brain regions is divided into the three groups of structural, effective and functional, s.t. the first one refers to the connectivity between nearby regions, while the second and third ones focus on the synchronization of oscillations of arbitrary located regions. Although EEG is not the best choice for analyzing functional and effective connectivity between brain regions due to its relatively poor spatial resolution, extracting its statistical features may be helpful in the analysis of synchronization of brain oscillations. In this paper, a novel framework for the prediction of seizure occurrence using EEG signals is proposed which utilizes the Granger causality approach in frequency domain to measure synchronization of EEG signals in the Inter-ictal and Pre-ictal time periods. Afterwards, a Logistic Regression classifier with Lasso regularization is used to discriminate the samples extracted from these two periods. At last, if a predefined number of consecutive samples are labled as Pre-ictals, a seizure occurrence alarm is issued. Experimental simulations on the CHB-MIT dataset resulted in 95.03% sensitivity and 0.14/hour false prediction rate, for 10min prediction horizon, which demonstrates effectiveness of our proposed method compared to the state-of-the-arts.
Full Research Paper
Sobhan Sheykhivand; Sehraneh Ghaemi
Volume 13, Issue 3 , October 2019, Pages 209-222
Abstract
The automatic classification of sleep stages is essential for the timely detection of disorders and sleep-related studies. In this paper, a single-channel EEG-based algorithm is used to automatically identify sleep stages using discrete wavelet transform and a hybrid model of ant colony optimizer and ...
Read More
The automatic classification of sleep stages is essential for the timely detection of disorders and sleep-related studies. In this paper, a single-channel EEG-based algorithm is used to automatically identify sleep stages using discrete wavelet transform and a hybrid model of ant colony optimizer and neural network based on RUSBoost. The signal is decomposed using a discrete wavelet transform into four levels and statistical properties of each level are calculated. To optimize and reduce the dimensions of feature vectors, hybrid model of ant colony optimizer algorithm and multi-layered neural network are used. Then ANOVA test is applied to validate the selected features. Finally, the classification is performed on RUSBoost, which provides an average of 90% classification accuracy for 2 to 6-class classification of different steps of sleep EEG. Suggesting that the proposed method has a higher degree of success in classifying sleep stages compared to the existing methods.
Full Research Paper
Mahla Dehtaghi Zadeh; Farid Saberi-Movahed; Mahdi Eftekhari
Volume 13, Issue 3 , October 2019, Pages 223-234
Abstract
DNA micro-array datasets play crucial role in machine learning and recognition of various kinds of cancer structures. Micro-array datasets are typically characterized by the high number of features and the small number of samples. Such problems may result in overfitting and low prediction accuracy of ...
Read More
DNA micro-array datasets play crucial role in machine learning and recognition of various kinds of cancer structures. Micro-array datasets are typically characterized by the high number of features and the small number of samples. Such problems may result in overfitting and low prediction accuracy of classifiers due to the irrelevant features, and therefore, they are considered as a challenging task in machine learning. The direct way to deal with such challenges is dimensionality reduction of data. In this regard, feature selection method acts as an effective solution for dimensinality reduction and increasing efficiency of learning algorithms. In this paper, by using the concept of “the basis for the DNA micro-array datasets”, a new feature selection method is introduced. To be more specific, rather than utilizing the entire micro-array dataset for tackling the problem of feature selection, a basis that is a muchmore smaller subset of the micro-array dataset is used. This method is based on subspace learning and matrix factorization. Finally, by making use of the DNA micro-array datasets, the effectiveness of the proposed method is evaluated, and the obtained results are compared with some state-of-the-art supervised feature selection methods.
Full Research Paper
Amir Babaoghli; Hadi Soltanizadeh
Volume 13, Issue 3 , October 2019, Pages 235-246
Abstract
Diseases associated with the retina and macula of the eye, causing permanent loss of vision or a great deal of loss of vision in people, leads to a decrease in the quality of life and a lot of problems in daily life. For this reason, the timely and correct identification of these diseases and disorders ...
Read More
Diseases associated with the retina and macula of the eye, causing permanent loss of vision or a great deal of loss of vision in people, leads to a decrease in the quality of life and a lot of problems in daily life. For this reason, the timely and correct identification of these diseases and disorders has become important. The optical coherence tomography imaging method provides high precision in imaging and good information about the depth of the retina. This imaging technique is a great help in the accurate identification of macular-related diseases. Age-related macular degeneration is one of the most common retinal diseases. The purpose of this study is to design and implement a system that is reliable, fast and can detect the age-related macular degeneration by using optical coherence tomography image processing accurately and quickly. In these studies, histograms of orientational gradients and principal component analysis for extraction of features and AdaBoost ensemble classification method have been used to classify the data. The database used includes 269 patients and 115 healthy people. All three indicators of accuracy, sensitivity and specificity of the implemented system were measured at 100%.
Full Research Paper
Seyedeh Somayeh Naghibi; Ali Fallah; Ali Maleki; Farnaz Ghassemi
Volume 13, Issue 3 , October 2019, Pages 247-257
Abstract
The correct prediction of the optimal motor trajectory is necessary for movement rehabilitation and control systems such as functional electrical stimulation and robotic therapy. It seems that human reaching movements are composed of a set of submovements, each of which is a correction of the overall ...
Read More
The correct prediction of the optimal motor trajectory is necessary for movement rehabilitation and control systems such as functional electrical stimulation and robotic therapy. It seems that human reaching movements are composed of a set of submovements, each of which is a correction of the overall movement trajectory. Therefore, it is possible to interpret complex movements, learning, adaptability and other features of the motion control system using submovements. The purpose of this study is predicting and generating planar reaching movements using a realistic model similar to the actual mechanism of human movement and based on the submovement. The data used consists of different replications of four types of planar movement Performed by three healthy subjects. After the preprocessing and phasing, the movements decomposed to minimum-jerk submovement. In the next step, the training of three distinct neural networks was carried out to learn the submovement parameters including the amplitude, duration, and initiation time. Finally, the ANNs were combined to form a closed-loop model that generated accurate reaching movements based on the error correction. The target access rate for all predicted movements by the closed loop model was 100%. Also, the mean distance to the target, the VAF, and the mean MSE error between the predicted and main movement trajectory showed that the predicted movements are a good approximation of the main movements. The results showed that when trained neural networks with submovements, were placed in a closed loop model, they were able to predict proper submovements for complete access to targets due to the compensation of propagated errors from the previous steps. The results of this study can be used to improve motor rehabilitation methods.
Full Research Paper
Malihe Molaie; Reza Aghaeizadeh Zoroofi
Volume 13, Issue 3 , October 2019, Pages 259-271
Abstract
Quantifying and modeling of the skeletal muscles can lead to an easier investigation of muscle diseases, specific mobility problems, and required simulations for the relevant surgeries. To this end, medical images should be segmented, firstly. In this research, thigh muscles segmentation is performed ...
Read More
Quantifying and modeling of the skeletal muscles can lead to an easier investigation of muscle diseases, specific mobility problems, and required simulations for the relevant surgeries. To this end, medical images should be segmented, firstly. In this research, thigh muscles segmentation is performed in CT images, since these muscles play a critical role in walking and balancing the body. To this aim, a multi-atlas method is used which is an improvement of the hierarchical multi-atlas method in the previous work. In this method, the muscles region is extracted automatically from the other tissues using FRFCM (Fast and Robust Fuzzy C-Means Clustering) method after the preprocessing stage. This muscle binary mask and the improved mask are used in the multi-atlas method for individual muscle segmentation. The proposed method is implemented using 20 CT data sets consisting of 12 female and 8 male subjects. The results show a less consumed computational time than the hierarchical multi-atlas method. The average computational time required for the muscles segmentation using the proposed method is 24 seconds and for the hierarchical multi-atlas method is 71 seconds per one slice of each case. Therefore, the proposed method reduces the implementation time by a rough factor of three. The means of the Dice similarity coefficient for the proposed method with improved muscle mask and for the hierarchical multi-atlas method are 86.58±7.69 and 83.07±8.26, respectively. The means of the precision and sensitivity for our method are 89.78±9.6 and 84.63±9.25, and for the hierarchical multi-atlas method are 88.85±12.04 and 78.04±10.88. Consequently, this method has better results based on the Dice similarity coefficient, precision, and sensitivity metrics.
Full Research Paper
Samira Rezvani Ardakani; Sajad Mohammad-Ali-Nezhad; Reza Ghasemi
Volume 13, Issue 3 , October 2019, Pages 273-289
Abstract
Epilepsy is one of the most important neurological disorders in the world. In order to suppress epileptic seizures, various control algorithms have been used. Time to control and reduce attacks and robustness of the controller against variations of pathologic parameters and unwanted oscillations are ...
Read More
Epilepsy is one of the most important neurological disorders in the world. In order to suppress epileptic seizures, various control algorithms have been used. Time to control and reduce attacks and robustness of the controller against variations of pathologic parameters and unwanted oscillations are important to control epileptic seizure. In order to consider these requirements and considering that one of the methods used to suppress epileptic waves is the change in mean soma (electric) potential of the excitatory neurons, this paper applies a fixed-time integral super twisting sliding mode controller to the combination of cortical and optogenetic models. First, the ion current produced in ion channels in optogenetic method is applied to the state variable of the mean electric potential of the excitatory neurons of the cortical model and the cortical and optogenetic models are combined and the controlled voltage applied to the system is applied to neurons of the epileptic zone of the brain as optic photons via the optogenetic model. Then, the mentioned controller is applied to the hybrid model so that the healthy model is tracked by the epileptic model in a fixed time. Finally, using the fixed-time integral super twisting sliding mode controller, the convergence error of the epileptic state to the healthy state has become zero. The amplitude of the control signal is reduced compared to the classic sliding mode control and technical problems and unwanted oscillations which are the shortcomings of the classic sliding mode controller are resolved.
