Non-stationarity is often observed in Geographic datasets. One way to explain non-stationarity is to think of it as a hidden "local knowledge" that varies across space. It is inherently difficult to model such data as models built for one region do not necessarily fit another area as the local knowledge could be different. A solution for this problem is to construct multiple local models at various locations, with each local model accounting for a sub-region within which the data remains relatively stationary. However, this approach is sensitive to the size of data, as the local models are only trained from a subset of observations from a particular region. In this paper, we present a novel approach that addresses this problem by aggregating spatially similar sub-regions into relatively large partitions. Our insight is that although local knowledge shifts over space, it is possible for multiple regions to share the same local knowledge. Data from these regions can be aggregated to train a more accurate model. Experiments show that this method can handle non-stationary and outperforms when the dataset is relatively small.

This study presents a comprehensive survey on mixed impulse and Gaussian denoising filters which are applied to an image in order to gauge the effects of this type of noise combination and to then determine optimal ways that can overcome such effects. The random noise model considered in this survey is the combined effect of impulse (salt and pepper) and Gaussian noise. After describing the noise models, the denoising filters which are applied to the images are classified and explained according to their design structure, the type of filters they use, the noise level they could overcome, and the limitations they face. This survey covers all related denoising methods and provides an assessment of the strengths and practical limitations of the different classes of denoising filters.

This study utilises a deep convolutional neural network (CNN) implementing regularisation and batch normalisation for the removal of mixed, random, impulse, and Gaussian noise of various levels from digital images. This deep CNN achieves minimal loss of detail and yet yields an optimal estimation of structural metrics when dealing with both known and unknown noise mixtures. Moreover, a comprehensive comparison of denoising filters through the use of different structural metrics is provided to highlight the merits of the proposed approach. Optimal denoising results were obtained by using a 20‐layer network with 40 × 40 patches trained on 400 180 × 180 images from the Berkeley segmentation data set (BSD) and tested on the BSD100 data set and an additional 12 images of general interest to the research community. The comparative results provide credence to the merits of the proposed filter and the comprehensive assessment of results highlights the novelty and performance of this CNN‐based approach.

Chest X-ray radiography is one of the earliest medical imaging technologies and remains one of the most widely-used for diagnosis, screening, and treatment follow up of diseases related to lungs and heart. The literature in this field of research reports many interesting studies dealing with the challenging tasks of bone suppression and organ segmentation but performed separately, limiting any learning that comes with the consolidation of parameters that could optimize both processes. This study, and for the first time, introduces a multitask deep learning model that generates simultaneously the bone-suppressed image and the organ-segmented image, enhancing the accuracy of tasks, minimizing the number of parameters needed by the model and optimizing the processing time, all by exploiting the interplay between the network parameters to benefit the performance of both tasks. The architectural design of this model, which relies on a conditional generative adversarial network, reveals the process on how the wellestablished pix2pix network (image-to-image network) is modified to fit the need for multitasking and extending it to the new image-to-images architecture. The developed source code of this multitask model is shared publicly on Github as the first attempt for providing the two-task pix2pix extension, a supervised/paired/aligned/registered image-to-images translation which would be useful in many multitask applications. Dilated convolutions are also used to improve the results through a more effective receptive field assessment. The comparison with state-of-the-art al-gorithms along with ablation study and a demonstration video1 are provided to evaluate the efficacy and gauge the merits of the proposed approach.

Alzheimer's disease (AD) is the most prevalent neurodegenerative disease that is progressive and can be characterized mostly by neuronal atrophy, amyloid deposition, accompanied by cognition, behavioral and psychological deficits. In the recent decade, a variety of machine learning algorithms have been explored and used for AD diagnosis, focusing on its subtle prodromal stage of mild cognitive impairment (MCI) to assess essential features that characterize its early manifestation and to plan for early treatment. However, diagnosis of early MCI (EMCI) remains most challenging as it is extremely difficult to delineate from cognitively normal controls (CN), and as a consequence, most of the classification algorithms for these two groups are mixed with low classification accuracy results. In this study, a machine learning approach based on deep neural network (DNN) has been proposed in order to detect AD in its early stage using multimodal imaging, combining magnetic resonance imaging (MRI), positron emission tomography (PET) and standard neuropsychological test scores. The proposed approach makes use of the optimization method of Adam to update the learning weights in order to improve its accuracy. The algorithm is able to classify cognitively normal control group from EMCI with an unprecedented accuracy of 84.0%. Although the focus here is distinguishing the two groups of CN and EMCI for early diagnosis and treatment planning, this study also shows how the proposed deep learning algorithm can be extended for multiclass classification involving CN and all the stages of EMCI, late MCI (LMCI) and AD.

Studies on progression and prediction of Alzheimer's disease (AD) at multiple time points are less prevalent than the standard cross-sectional studies, which have been around for over two decades. The reason is that longitudinal datasets have not been readily available until recently, therefore predictive analysis of AD is relatively new to the scene. Even though some new lines of research have started to focus on regression modeling of the patient's pattern change in the future, they neglect the fact that the progression curves for different initial profiles at the baseline are fundamentally different from each other. These trends can hardly be modeled using a general regressor no matter how complex and nonlinear the model is. Therefore, encoding the disease progression for patients with different disease severity using separate regressors may achieve results that are more reliable. Following this assertion, we propose a model that in its first step classifies the subjects in the baseline data into four groups of Early Mild Cognitive Impairment (EMCI), Late Mild Cognitive Impairment (LMCI), and Cognitively Normal Controls (NC). In the second stage, we trained profile specific regressors to estimate Mini-Mental State Examination (MMSE) scores in multiple future time-points up to 36 months ahead of time. To validate our model, we perform a set of experiments on ADNI baseline MRI, FDG-PET, and cerebrospinal fluid (CSF) data from 1,458 subjects, among them 333 AD patients, 529 LMCI, 255 EMCI, and 341 NC. Experimental results demonstrate the effectiveness of the proposed method in terms of Root Mean Square Error (RMSE) exhibiting an average of 1.12 decrease in the MMSE questionnaire scale range of [0 30] over comparable RMSE scores obtained with state-of-the-art regression kernels.

This study introduces an image denoising method in the presence of combined speckle and Gaussian noise. The dual‐tree complex wavelet transform is applied to the image in order to obtain specific coefficients characterising these types of noise. Then, these extracted coefficients are removed by thresholding and an inverse wavelet transform is applied in order to obtain the reconstructed image. A comparison between the dual‐tree and standard wavelet‐based denoising filters is provided on the basis of different structural metrics. Finally, in order to remove any remaining noise, a spatial denoising filter is applied to the image. The results obtained on medical ultrasound images corrupted by this noise combination support the authors’ assertion on the method's resilience to the combined effects of speckle and Gaussian noise, and are compared to well‐known and most effective speckle–Gaussian denoising filters.

Over the past few years, several approaches have been proposed to assist in the early diagnosis of Alzheimer's disease (AD) and its prodromal stage of mild cognitive impairment (MCI). Using multimodal biomarkers for this high-dimensional classification problem, the widely used algorithms include Support Vector Machines (SVM), Sparse Representation-based classification (SRC), Deep Belief Networks (DBN) and Random Forest (RF). These widely used algorithms continue to yield unsatisfactory performance for delineating the MCI participants from the cognitively normal control (CN) group. A novel Gaussian discriminant analysis-based algorithm is thus introduced to achieve a more effective and accurate classification performance than the aforementioned state-of-the-art algorithms. This study makes use of magnetic resonance imaging (MRI) data uniquely as input to two separate high-dimensional decision spaces that reflect the structural measures of the two brain hemispheres. The data used include 190 CN, 305 MCI and 133 AD subjects as part of the AD Big Data DREAM Challenge #1. Using 80% data for a 10-fold cross-validation, the proposed algorithm achieved an average F1 score of 95.89% and an accuracy of 96.54% for discriminating AD from CN; and more importantly, an average F1 score of 92.08% and an accuracy of 90.26% for discriminating MCI from CN. Then, a true test was implemented on the remaining 20% held-out test data. For discriminating MCI from CN, an accuracy of 80.61%, a sensitivity of 81.97% and a specificity of 78.38% were obtained. These results show significant improvement over existing algorithms for discriminating the subtle differences between MCI participants and the CN group.

This study introduces a robust edge detection method that relies on an integrated process for denoising images in the presence of high impulse noise. This process is shown to be resilient to impulse (or salt and pepper) noise even under high intensity levels. The proposed switching adaptive median and fixed weighted mean filter (SAMFWMF) is shown to yield optimal edge detection and edge detail preservation, an outcome we validate through high correlation, structural similarity index and peak signal to noise ratio measures. For comparative purposes, a comprehensive analysis of other denoising filters is provided based on these various validation metrics. The nonmaximum suppression method and new edge following maximum-sequence are two techniques used to track the edges and overcome edge discontinuities and noisy pixels, especially in the presence of high-intensity noise levels. After applying predefined thresholds to the grayscale image and thus obtaining a binary image, several morphological operations are used to remove the unwanted edges and noisy pixels, perform edge thinning to ultimately provide the desired edge connectivity which results in an optimal edge detection method. The obtained results are compared to other existing state-of-the-art denoising filters and other edge detection methods in support of our assertion that the proposed method is resilient to impulse noise even under high-intensity levels.

This study introduces a novel Gaussian discriminant analysis (GDA)-based computer aided diagnosis (CAD) system using structural magnetic resonance imaging (MRI) data uniquely as input for screening different stages of Alzheimers disease (AD) involving its prodromal stage of mild cognitive impairment (MCI) in relation to the cognitive normal control group (CN). Taking advantage of multiple modalities of biomarkers, over the past few years, several machine learning-based CAD approaches have been proposed to address this high-dimensional classification problem. This study presents a novel GDA-based CAD system on the basis of a tenfold cross validation and a held-out test data set. Subjects considered in this study included 187 CN, 301 MCI, and 131 AD subjects from the Alzheimers Disease Neuroimaging Initiative (ADNI) database. In the tenfold cross validation, the proposed system achieved an average F1 score of 97.20%, accuracy of 96.00%, sensitivity of 99.14%, and specificity of 88.67% for discriminating together the MCI and AD groups from the CN group; and an average F1 score of 79.82%, accuracy of 87.43%, sensitivity of 79.09%, and specificity of 91.25% for discriminating AD from MCI. By testing on the held-out test data, for discriminating MCI and AD from CN, an accuracy of 93.28%, a sensitivity of 98.78%, and a specificity of 81.08% were obtained. These results also show that by separating left and right hemispheres of the brain into two decisional spaces, and then combining their outputs, the GDA-based CAD system demonstrates a high potential for clinical application.

This paper introduces the classification of patterns extracted from different types of interictal epileptiform discharges (IEDs) that includes interictal spike (IS), spike and slow wave complex (SSC), and repetitive spikes and slow wave complex (RSS)), using the partial directed coherence (PDC) analysis. The PDC analysis estimates the intensity and direction of propagation from neural activities generated in the cerebral cortex, and analyzes the coefficients obtained from employing multivariate autoregressive model (MVAR). Features extracted by using PDC are transformed into binary matrices by using surrogate data testing with a 0.05 significance level. The significant propagations are represented as 1 in the binary matrix and 0 otherwise. Binary matrices are converted into binary vectors. These vectors are then selected as the inputs of a multilayer Perceptron (MLP) neural network. The first classifier is trained to distinguish between 2 types of IEDs and tenfold cross validation is implemented to evaluate the system. The performance of the classifier was evaluated, where it achieved the highest F1 score of 100.00% when performed on IS vs RSS and 96.67% on IS vs CSS. The average F1 score of the first classifier obtained was 91.11%. The second classifier was trained to perform all types of IEDs classifications. The classifier yielded an overall accuracy of 86.67% with the highest achieved F1 score of 90.00%. Both classifiers were able to detect and classify different types of IEDs when using the features extracted from PDC with a very high performance.

Neuroimaging is an important research platform that can be very useful for eliciting new understanding on the complicated pathogenesis between genetics and disease phenotypes. Due to the extremely high dimensionality of image and genetic data, and considering the potential joint effect of genetic variants, multivariate techniques have been examined to detect Alzheimers disease (AD) related genetic variants expressed through single-nucleotide polymorphisms (SNPs). However, the image features used in support of those methods are not immediately related to the disease, and the detected genetic markers may not be related to AD. In this study, we propose an ensemble model based framework for firstly extracting 50 region-based image features whose values are predicted by base learners trained on raw neuroimaging morphological variables. This task is followed by performing sparse Partial Least Squares regression (sPLS) method on the extracted 50 AD related image features and pre-selected 1508 SNPs to detect the significant SNPs associated with the extracted image features. Instead of modeling a direct link between genetic variants and disease label, we captured disease information indirectly.

Patterns of interictal epileptiform activities, such as sharp waves, spikes, spike-wave complexes and polyspike-wave complexes are explored in the recorded electroencephalograms (EEG) to gauge the different functional connectivity dynamics and to assess how they could be affected by the type of a seizure. Connectivity measures were represented by the phase synchronization among scalp electrodes that were obtained by adopting a nonlinear data-driven method. These interictal epileptic activities were investigated using a graph theory analysis. The connectivity maps were compared by considering the number of connections in four main brain regions (anterior region, posterior region, left hemisphere and right hemisphere). Results revealed interesting and different network topology for the connectivity maps. Besides, a relationship between the connectivity patterns of the recorded epileptic activities and the types of seizures was observed. This relationship was statistically confirmed by analysis of variance (ANOVA) that denoted a significant difference among connectivity patterns of sharp waves and spike activities, which were seen in focal epilepsy, in contrast to the spike-wave and polyspike-wave complexes that were associated with generalized epilepsy (P-value = 0). These results augment the prospects for diagnosis and enhance the recognition of the disease type via EEG-based connectivity maps.

Typically, neuropsychological testing helps medical experts situate a given patient in continuum of the Alzheimer's disease (AD) spectrum, especially in the continuum between cognitively normal controls (CN) and the prodromal stage of mild cognitive impairment (MCI). As a well-known early symptom, some linguistic complexity changes of language have been associated with the progression of AD. Currently, the investigations of linguistic manifestations of MCI do mostly rely on manual analysis, yet linguistic complexity-based neuropsychological assessment cannot be effectively attained by using manual approaches. Taking advantage of the existing natural language processing (NLP) techniques to extract key features in linguistic complexity changes associated with the progression of MCI, this study develops an NLP-oriented computerized neuropsychological assessment to automatically analyze the distinguishing characteristics of data in the MCI group versus those in cognitively normal control (CN) group. By using the transcripts of news conferences from President Ronald Reagan, who was diagnosed with Alzheimer's disease in 1994, and President Dwight D. Eisenhower (DE), and President George H. W. Bush (GB), who have no known diagnosis of AD, this study found the significantly different patterns of the linguistic complexity changes between CN and MCI subjects, which can be applied into the diagnosis of MCI.

Over the past few years, several approaches have been proposed to assist in the early diagnosis of Alzheimer's disease (AD) and its prodromal stage of mild cognitive impairment (MCI). For solving this high dimensional classification problem, the widely used algorithm remains to be Support Vector Machines (SVM). But due to the high variance of the data, the classification performance of SVM remains unsatisfactory, especially for delineating the MCI group from the cognitively normal control (CN) group. This study introduces a novel algorithm based on the Gaussian discriminant analysis (GDA) for a more effective and accurate classification performance. Subjects considered in this study included 190 CN, 305 MCI, and 133 AD subjects. Using 75% of the data as the training set with a tenfold cross validation, the proposed algorithm achieved an average accuracy of 94.17%, a sensitivity of 93.00%, and a specificity of 95.00% for discriminating AD from CN; and an average accuracy of 84.86%, a sensitivity of 84.78%, and a specificity of 85.00% for discriminating MCI from CN. Then a true test was implemented for the remaining 25% data, for discriminating specifically MCI from CN, resulting in an accuracy of 82.20%, a sensitivity of 83.10%, and a specificity of 80.85%. As revealed through the literature, these results involving the delineation of the MCI group from CN could be considered as the best classification performance obtained so far. This study also shows that by separating left and right hemispheres of the brain into two decision spaces, then combining the results of these two spaces, the classification performance can be improved significantly; an assertion proven in this study.

Background: Deposition of beta amyloid protein (Aβ) is known to be an early event that is closely associated with the pathogenesis of Alzheimer's disease (AD), along with related downstream events such as neuronal loss, neurofibrillary tangles, cortical thinning and cognitive deficits. APOE e4 allele (E4) is also known to be associated with increased risk for AD. Objectives: The goal of this study is to examine the association of Aβ deposition to cortical thickness (CoTh), in healthy control (CN), early MCI (EMCI), late MCI (LMCI) and AD stages by controlling for E4 load, both in regional and hemispheric levels, and to interpret patterns of different brain regions based on their correlation performance among the four groups. Methods: We analyzed Amyloid PET Scan, Volumetric MRI (CoTh) data from participants in the ADNIGO/ADNI2 cohort whose APOE gene information are available. Statistical analysis includes Pearson partial correlations, Analysis of Covariance (ANCOVA) with post-hoc Tukey HSD. Complete-linkage hierarchical clustering analysis was further performed to group brain regions based on their significant correlation performance. Results: 25 out of 68 regions showed significant correlation of Aβ load and CoTh at least in one diagnostic group. Furthermore, 6 main clusters were recognized based on the performance patterns of those 25 regions across 4 diagnosis groups. Conclusion: Our major finding clustered the cortical regions into 2 general groups, positive correlation in CN or AD, and negative correlation in EMCI and/or LMCI, and 6 more specific groups were then recognized, confirming the interplay between of Aβ and CoTh in the different stages of the disease.

This study introduces a novel learning algorithm for spiking neurons, called CCDS, which is able to learn and reproduce arbitrary spike patterns in a supervised fashion allowing the processing of spatiotemporal information encoded in the precise timing of spikes. Unlike the Remote Supervised Method (ReSuMe), synapse delays and axonal delays in CCDS are variants which are modulated together with weights during learning. The CCDS rule is both biologically plausible and computationally efficient. The properties of this learning rule are investigated extensively through experimental evaluations in terms of reliability, adaptive learning performance, generality to different neuron models, learning in the presence of noise, effects of its learning parameters and classification performance. Results presented show that the CCDS learning method achieves learning accuracy and learning speed comparable with ReSuMe, but improves classification accuracy when compared to both the Spike Pattern Association Neuron (SPAN) learning rule and the Tempotron learning rule. The merit of CCDS rule is further validated on a practical example involving the automated detection of interictal spikes in EEG records of patients with epilepsy. Results again show that with proper encoding, the CCDS rule achieves good recognition performance.

Understanding pathways of neurological disorders requires extensive research on both structural and functional characteristics of the brain. Graph theoretical analysis of functional connectivity networks highlight the dynamics of communication among brain neurons. Resembling brain to a Networked Control System, describes seizure, epileptic episodes, as a response to a set of triggers. This unknown set attempts to destabilize the system and so referred as cyberattacks in network terminology. The study investigates the dynamics of brain networks under the attack of epileptic seizures focusing on the network parameters variation for pre- inter- and post-attack durations. Statistical analysis of the Scalp EEG-based connectivity networks found significantly (p <; 0.05) higher global efficiency and lower clustering coefficient during the attack compared to those of before attack onset and system reinstatement. Study did not support the existence of statistically significant difference in small world index for inter-attack duration, however network tends to scatter from small word model.

BACKGROUND: The lives of half a million children in the United States are severely affected due to the alterations in their functional and mental abilities which epilepsy causes. This study aims to introduce a novel decision support system for the diagnosis of pediatric epilepsy based on scalp EEG data in a clinical environment. METHODS: A new time varying approach for constructing functional connectivity networks (FCNs) of 18 subjects (7 subjects from pediatric control (PC) group and 11 subjects from pediatric epilepsy (PE) group) is implemented by moving a window with overlap to split the EEG signals into a total of 445 multi-channel EEG segments (91 for PC and 354 for PE) and finding the hypothetical functional connectivity strengths among EEG channels. FCNs are then mapped into the form of undirected graphs and subjected to extraction of graph theory based features. An unsupervised labeling technique based on Gaussian mixtures model (GMM) is then used to delineate the pediatric epilepsy group from the control group. RESULTS: The study results show the existence of a statistically significant difference (p < 0.0001) between the mean FCNs of PC and PE groups. The system was able to diagnose pediatric epilepsy subjects with the accuracy of 88.8% with 81.8% sensitivity and 100% specificity purely based on exploration of associations among brain cortical regions and without a priori knowledge of diagnosis. CONCLUSIONS: The current study created the potential of diagnosing epilepsy without need for long EEG recording session and time-consuming visual inspection as conventionally employed.

BACKGROUND: Intracranial volume (ICV) is an important normalization measure used in morphometric analyses to correct for head size in studies of Alzheimer Disease (AD). Inaccurate ICV estimation could introduce bias in the outcome. The current study provides a decision aid in defining protocols for ICV estimation in patients with Alzheimer disease in terms of sampling frequencies that can be optimally used on the volumetric MRI data, and the type of software most suitable for use in estimating the ICV measure. METHODS: Two groups of 22 subjects are considered, including adult controls (AC) and patients with Alzheimer Disease (AD). Reference measurements were calculated for each subject by manually tracing intracranial cavity by the means of visual inspection. The reliability of reference measurements were assured through intra- and inter- variation analyses. Three publicly well-known software packages (Freesurfer, FSL, and SPM) were examined in their ability to automatically estimate ICV across the groups. RESULTS: Analysis of the results supported the significant effect of estimation method, gender, cognitive condition of the subject and the interaction among method and cognitive condition factors in the measured ICV. Results on sub-sampling studies with a 95% confidence showed that in order to keep the accuracy of the interleaved slice sampling protocol above 99%, the sampling period cannot exceed 20 millimeters for AC and 15 millimeters for AD. Freesurfer showed promising estimates for both adult groups. However SPM showed more consistency in its ICV estimation over the different phases of the study. CONCLUSIONS: This study emphasized the importance in selecting the appropriate protocol, the choice of the sampling period in the manual estimation of ICV and selection of suitable software for the automated estimation of ICV. The current study serves as an initial framework for establishing an appropriate protocol in both manual and automatic ICV estimations with different subject populations.

The concept of 2-D barcodes is of great relevance for use in wireless data transmission between handheld electronic devices. In a typical setup, any file on a cell phone, for example, can be transferred to a second cell phone through a series of images on the LCD which are then captured and decoded through the camera of the second cell phone. In this study, a new approach for data modulation in 2-D barcodes is introduced, and its performance is evaluated in comparison to other standard methods of barcode modulation. In this new approach, orthogonal frequency-division multiplexing (OFDM) modulation is used together with differential phase shift keying (DPSK) over adjacent frequency domain elements. A specific aim of this study is to establish a system that is proven tolerant to camera movements, picture blur, and light leakage within neighboring pixels of an LCD.

Most of the computer users with ocular aberrations such as myopia, hyperopia and other high-order aberrations are subject to visual blurring, which may impede the efficient interactions with the computers. Conventional methods used to counter visual blurring include spectacles and contact lenses. In this paper, we introduce an image preprocessing method that is designed to counteract the visual blurring caused by the aberration of the eye. In this method, the presented images are preprocessed by performing personalised compensation according to the ocular aberration of the computer user. In order to overcome the mismatch between the aberration used to generate the precompensation and the actual aberrations at the time of viewing, dynamic ocular aberrations are derived from the resizing of the initial aberration data measured by the wavefront analyzer. The dynamic ocular aberrations are used to update the image precompensation in real time. Results of human subject experiment show that the image recognition accuracy was significantly increased after the dynamic precompensation was applied. Subjective impressions from the participants confirmed the effectiveness of the method.

This study introduces a new Generalized Leaky Integrate-and-Fire (GLIF) neuron model with variable leaking resistor and bias current in order to reproduce accurately the membrane voltage dynamics of a biological neuron. The accuracy of this model is ensured by adjusting its parameters to the statistical properties of the Hodgkin-Huxley model outputs; while the speed is enhanced by introducing a Generalized Exponential Moving Average method that converts the parameterized kernel functions into pre-calculated lookup tables based on an analytic solution of the dynamic equations of the GLIF model.

Previous research found that the pupil diameter (PD) can be an indication of affective state, but this approach to the detection of the affective state of a computer user has not been investigated fully. We propose a new affective sensing approach to evaluate the computer user's affective states as they transition from "relaxation” to "stress,” through processing the PD signal. Wavelet denoising and Kalman filtering were used to preprocess the PD signal. Then, three features were extracted from it and five classification algorithms were used to evaluate the overall performance of the identification of "stress” states in the computer users, achieving an average accuracy of 83.16 percent, with the highest accuracy of 84.21 percent reached with a Multilayer Perceptron and a Naive Bayes classifier. The Galvanic Skin Response (GSR) signal was also analyzed to study the comparative efficiency of affective sensing through the PD signal. We compared the discriminating power of the three features derived from the preprocessed PD signal to three features derived from the preprocessed GSR signal in terms of their Receiver Operating Characteristic curves. The results confirm that the PD signal should be considered a powerful physiological factor to involve in future automated affective classification systems for human-computer interaction.

A new thermal imaging framework with unique feature extraction and similarity measurements for face recognition is presented. The research premise is to design specialized algorithms that would extract vasculature information, create a thermal facial signature and identify the individual. The proposed algorithm is fully integrated and consolidates the critical steps of feature extraction through the use of morphological operators, registration using the Linear Image Registration Tool and matching through unique similarity measures designed for this task. The novel approach at developing a thermal signature template using four images taken at various instants of time ensured that unforeseen changes in the vasculature over time did not affect the biometric matching process as the authentication process relied only on consistent thermal features. Thirteen subjects were used for testing the developed technique on an in-house thermal imaging system. The matching using the similarity measures showed an average accuracy of 88.46% for skeletonized signatures and 90.39% for anisotropically diffused signatures. The highly accurate results obtained in the matching process clearly demonstrate the ability of the thermal infrared system to extend in application to other thermal imaging based systems. Empirical results applying this approach to an existing database of thermal images proves this assertion.

Visual distortion and blurring impede the efficient interaction between computers and their users. Visual problems can be caused by eye diseases, severe refractive errors or combinations of both. Several image enhancement methods based on contrast sensitivity have been used to help people with eye diseases (e.g., age-related macular degeneration and cataracts), whereas few methods have been designed for people with severe refractive errors. This paper describes a new pre-compensation method to counter the visual blurring caused by the severe refractive errors of a specific computer user. It preprocesses the pictorial information through dynamic pre-compensation in advance, aiming to present customized images on the basis of the ocular aberrations of the specific computer user. The new method improves the previous static pre-compensation method by updating the aberration data according to pupil size variations, in real-time. The real-time aberration data enable us to generate better suited pre-compensated images, as the pre-compensation model is updated dynamically. An empirical study was conducted to evaluate the efficiency of the new pre-compensation method, through an icon recognition test. From the results of statistical analysis, we found that participants achieved significantly higher accuracy levels in recognizing the icons with dynamic pre-compensation, than when viewing the original icons. The accuracy is also significantly boosted when the icons were processed with dynamic pre-compensation method, in comparison with the previous static pre-compensation method.

This study evaluates the sensitivity, specificity and accuracy in associating scalp EEG to either control or epileptic patients by means of artificial neural networks (ANNs) and support vector machines (SVMs). A confluence of frequency and temporal parameters are extracted from the EEG to serve as input features to well-configured ANN and SVM networks. Through these classification results, we thus can infer the occurrence of high-risk (epileptic) as well as low risk (control) patients for potential follow up procedures.

This study describes a new 3-D liver segmentation method in support of the selective internal radiation treatment as a treatment for liver tumors. This 3-D segmentation is based on coupling a modified k-means segmentation method with a special localized contouring algorithm. In the segmentation process, five separate regions are identified on the computerized tomography image frames. The merit of the proposed method lays in its potential to provide fast and accurate liver segmentation and 3-D rendering as well as in delineating tumor region(s), all with minimal user interaction. Leveraging of multicore platforms is shown to speed up the processing of medical images considerably, making this method more suitable in clinical settings. Experiments were performed to assess the effect of parallelization using up to 442 slices. Empirical results, using a single workstation, show a reduction in processing time from 4.5 h to almost 1 h for a 78% gain. Most important is the accuracy achieved in estimating the volumes of the liver and tumor region(s), yielding an average error of less than 2% in volume estimation over volumes generated on the basis of the current manually guided segmentation processes. Results were assessed using the analysis of variance statistical analysis.

Resource virtualization technologies have recently increased in popularity. The emergence of cloud computing, which requires provisioning isolated environments on shared resources, is one reason for this. Virtualization adds flexibility in terms of resource provisioning, but it can impact application performance. In this work, we analyze the performance of medical image processing and computational fluid dynamics applications when run on virtualized resources. We then apply the observed performance characteristics to a performance prediction model. We measure the impact of virtualization by performing several benchmarks on virtualized and non-virtualized resources. We evaluate the accuracy of the performance prediction model in this environment. We find that virtualization can slow down some applications by more than 200%, but usually the performance impact is below 15%. The overhead itself is predictable if general application characteristics are known. Execution time in a virtual environment can be predicted to within 13% using a simple mathematical prediction model.

Hispanics have the highest growth rates among all groups in the U.S., yet they remain considerably underrepresented in computing careers and in the numbers who obtain advanced degrees. Hispanics constituted about 7% of undergraduate computer science and computer engineering graduates and 1% of doctoral graduates in 2007--2008. The small number of Hispanic faculty, combined with the lack of Hispanic role models and mentors, perpetuates a troublesome cycle of underrepresentation in STEM fields. In 2004, seven Hispanic-Serving Institutions (HSIs) formed the Computing Alliance of Hispanic-Serving Institutions (CAHSI) to consolidate their strengths, resources, and concerns with the aim of increasing the number of Hispanics who pursue and complete baccalaureate and advanced degrees in computing areas. To address barriers that hinder students from advancing, CAHSI defined a number of initiatives, based on programs that produced promising results at one or more institutions. These included the following: a CS-0 course that focuses on adoption of a three-unit pre-CS course that uses graphics and animation to engage and prepare students who have no prior experience in computing; a peer mentoring strategy that provides an active, collaborative learning experience for students while creating leadership roles for undergraduates; an undergraduate and graduate student research model that emphasizes the deliberate and intentional development of technical, team, and professional skills and knowledge required for research and cooperative work; and a mentoring framework for engaging undergraduates in experiences and activities that prepare them for graduate studies and onto the professoriate. CAHSI plays a critical role in evaluating, documenting, and disseminating effective practices that achieve its mission. This paper provides an overview of CAHSI initiatives and describes how each addresses causes of underrepresentation of Hispanics in computing. In addition, it describes the evaluation and assessment of the initiatives and presents the results that support CAHSI’s claim of their effectiveness.

This study describes the performance results on testing MatLab applications using the parallel computing and the distributed computing toolboxes under different platforms with different hardware and operating systems. Each trial was executed keeping the hardware fixed and changing the operating system to obtain unbiased results. To standardize the benchmarking test, Fast Fourier Transform (FFT), discrete cosine transform (DCT), edge detection and matrix multiplication algorithms were executed. The results show that the leveraging of multicore platforms can speed up considerably the processing of images through the use of parallel computing tools in MatLab. Two different system hardware platforms (systems 1 and 2) were used in a series of experiments. Four rounds of experiments were performed benchmarking the FFT algorithm using the parallel tool box, by changing system platform, number of workers, image size and number of images. The results of the ANOVA test suggest that although there is no statistical significance on the factor represented by the operating system (OS) on system 1, the OS plays a significant roll on system 2. Moreover, on both systems there is statistical significance on the factors represented by the number of workers utilized and the number of images processed, yielding more than a 500% performance increase by using 8 MatLab workers on a dual quad-core machine.

Spatialized ("3D") audio may be useful as an alternative sensory channel to provide blind individuals with relevant spatial information in the physical world or while interacting with computers. However, an important limitation of this approach is the lower spatial resolution achievable through sound localization. While this limitation is widely acknowledged, few empirical evaluations of the sound localization achievable through audio spatialization techniques have been performed, particularly with respect to elevation localization. We performed such an empirical study and found quantitative confirmation that the localization accuracy deteriorates as the virtual sound position is set farther above or below the ear (height) level. This information may be valuable to HCI designers planning to use 3D sound.

The emergence of data-intensive applications in medical and other fields of study has created the need for increased storage space. It is often the case that these data should not or cannot be placed in the same domain. The data usually requires considerable computational power to process, which has inspired the area of Grid computing. This study proposes a new methodology for sharing storage volumes across distributed domains securely and making them accessible to all members of a grid. The proposed system was designed with the intent to integrate with the Globus Toolkit, which gives it the advantage of having useful features that are commonly necessary for data intensive applications.

In this paper, we propose the possibility of using real-time pupil diameter measurements, which are provided by most Eye Gaze Tracking (EGT) systems utilized for computer input by many individuals with severe motor impairments, to obtain an on-line estimation of the quality of the interaction with the computer, as perceived by the user. Increased computational power available to process, in real-time, pupil diameter data could, in the future, enable an assessment of the level of frustration (stress) that an EGT user may experience when the quality of the interaction deteriorates. This would be a critical first step in the direction of enabling EGT systems to initiate a re-calibration process, under those circumstances.

In this paper, we present an approach to teaching induction that we call the "conceptual route" of teaching induction. Proofs by induction are central to the study of computer science and students come across them in many courses of their curricula. It is documented in the literature that in general students have difficulties with proofs by induction. Even though through the years some solutions were proposed to improve the situation, recent studies show that students are still having difficulties. Currently, proofs by induction take up too little of the computer science curriculum, and they are taught as a step-by-step procedure to be followed, which is not sufficient for students to gain conceptual understanding. In contrast, our approach aims at students' conceptual understanding by shifting their focus from the syntactic form of proofs by induction to their substance. The theoretical underpinning of our approach is an operationalization of the Induction Principle.

In this paper, we present statistical results from testing our precompensation algorithms with 20 human subjects. A factorial experiment was designed and tested to evaluate the significance that the method, icon size, and subject group have on the ability of users to identify icons. These results reinforce software and "artificial eye" test findings indicating that the methods used for precompensation provide a significant increase in retinal image quality for users that have visual aberrations present in the optical systems of their eyes. Significant interactions were also found that reveal circumstances under which the method may perform best.

Eye Gaze Tracking (EGT) systems allow individuals with motor disabilities to quickly move a screen cursor on a PC. However, there are limitations in the steadiness and the accuracy of cursor control and clicking capabilities they provide. On the other hand, a cursor control system to step the cursor up, down, left or right in response to voluntary contractions of specific facial muscles, developed by our group, provides steady and precise, albeit slow, cursor control, along with a reliable clicking mechanism. This system identifies muscle contractions by performing digital processing of the Electromyogram (EMG) signals generated by the facial muscles. Based on the complementary nature of the strengths of these two cursor control modalities we have developed an integrated EGT/EMG system in an attempt to consolidate the advantages of both input modalities. We have compared the selection accuracy and speed of an EGT-only cursor control implementation, our integrated EGT/EMG cursor control system and a standard handheld mouse in point-and click trials.

In this paper, we describe a multi-domain approach for enhancing text displayed on a computer screen for users with visual aberrations. This research is based on a priori knowledge of the user's visual aberration, as measured by a wavefront analyzer. With this information it is possible to generate text that, when displayed to this user, will counteract his/her visual aberration. The method described in this paper advances the development of techniques for providing such compensation by integrating spatial information in the image as a means to eliminate some of the shortcomings inherent in using display devices such as monitors or LCD panels.

The possibility of pre-compensating images in a computer display according to the visual aberrations previously assessed in an optical system (e.g., the computer user's eye) has been confirmed for a simple "artificial eye". This device has been constructed from optical components, which include a plano-convex lens, an adjustable aperture, and a Charged-Couple Device (CCD) array that mimics the retina of a real eye. While the CCD array allows for the inspection of the image as it would form on the retina of a real eye, its specular reflection does not allow the resulting "artificial eye" to be measured appropriately in a wavefront analyzer (a necessary pre-requisite for the image precompensation process). Therefore, an alternative, interchangeable CCD array covered with gray paint (i.e., disabled) was also created to provide the diffuse reflectivity that is presumed in the operation of the wavefront analyzer. Experiments with this system show that the visual aberrations in a properly characterized optical system can, in fact, be precompensated by the methods proposed by Alonso et al., [1]. These same experiments, however, reveal the need to adjust the precompensation method according to the effective pupil diameter in the system during viewing.

In fields such as bioinformatics, cytometry, geographic information systems, just to name a few, huge amount of data, often multidimensional in nature, has more than ever highlighted the need for new algorithms to reduce the computational requirements needed for data analysis and interpretation. In this study, we present a new unsupervised clustering algorithm /sub e/nsity-based adaptive window clustering algorithm, which reduces the computational load to /spl sim/ O(N) number of computations, making it more attractive and faster than current hierarchical algorithms. This method relies on weighting a dataset to grid points on a mesh, and identifies the density peaks by reducing low density points, ranking and correlation calculation. The adaptive windows used are a modification of the recently proposed k-windows clustering algorithm to shape the desired clusters. The new algorithm makes it easier for users to observe and analyze data for enhanced interpretation and improved real-world applications, especially in clinical practices.

Computer technologies, frequently employed for everyday tasks, often use Graphical User Interfaces (GUIs), presented through monitors or LCD displays. This type of visual interface is not well suited for users with refractive visual limitations, particularly when they are severe and not correctable by common means. In order to facilitate computer access for users with refractive deficiencies, an algorithm has been developed, using a priori knowledge of the visual aberration, to generate on-screen images that counter the effect of the aberration. When the user observes the screen displaying a pre-compensated image, the image perceived in the retina will be similar to the original image. The algorithm was tested by artificially introducing a spherical aberration in the field of view of 14 subjects, totaling 28 individual eyes. This use of pre-compensation improves the visual performance of the subjects with respect to that achieved with no compensation.

Motion estimation and compensation combined with wavelet transformation has been beneficially applied to improve video compression efficiency and accuracy. Based on the similar motion structures between the decomposed subimages, wavelet domain video compression can be improved by means of the simplifications in motion field refinement as well as in motion vector representation. Although the utilization of the motion field correlations between the same orientational subbands can achieve better compression performance than that between the approximation subbands, the computational load has to be traded off. Since most of the energies are concentrated in the low resolution subbands while decreased in the high resolution subbands, by considering both the accuracy and complexity, an efficient approach, called level-refined motion estimation and subband compensation (LRSC) method is proposed. It realizes low-entropy coding in the subbands, while keeping low computational load in motion estimation, thus to achieve both temporal compression quality and computational simplicity.

The method introduced in this article is a new approach to text compression that shows merit in a higher compression ratio, its ability to be integrated with existing utilities to augment their performance, and its resiliency to error introduced through bit corruption by limiting the error effect to only that word where the error occurred. The results introduced, based on a comprehensive evaluation of the proposed method with respect to existing utilities, demonstrate such merits and are provided in support of these assertions.

An approach to stereo feature matching is presented with the introduction of a similarity measure for evaluating and confirming a stereo match. The contributions of this study are reflected in (1) the development of a similarity measure which evaluates a stereo match based on feature locality and gray-level gradient associated with the feature; and (2) the use of a matching procedure that integrates local and global matching strategies based on matching first those features with the highest similarity measure among the set of all highest similarities found locally under confined search spaces, ensuring that each feature is matched with a high degree of certainty. A left-to-right and right-to-left consistency check is used for each feature to comply with the uniqueness constraint and to confirm if a potential match can be declared a correct match.

Describes a new feature-based stereo matching technique which exploits the Walsh transformation. The established matching strategy adopts and integrates the fundamental steps of the stereo vision problem: (a) detecting and locating feature points, (b) searching for potential matches, (c) validating a match through a global consistency check, and (d) determining the disparity of the matched feature points. The unique representation of stereo images into Walsh-based attributes unites the aforementioned steps into an integrated process which yields accurate disparity extraction. It is shown that the first and second Walsh attributes are used as operators approximating the first and second derivatives for the extraction and localization of feature points. The complete set of these attributes are then used as matching primitives contributing equally to the decision-making process and providing relevant information on both the characterization of a potential match and its validation through a consistency check. Computer results using images of varying complexities prove the soundness and the relatively fast processing time of this stereo matching technique.>

Recent research into different artificial neural network structures and topologies suggests the possibility of implementing a particular application. The goal is for the neural network to represent the input function in a natural manner. The authors describe such an implementation in the field of neurosurgical planning, where a set of neural networks represents the lesion to be treated as well as the different functional regions of the brain. It is shown that this neural network structure can actively and effectively assist in the surgical planning. Emphasis is on stereotactic radiosurgery, whereby a high dose of radiation is delivered to the lesion. This modality allows for extensive implementation of the neural network features in a natural way, using Gaussian potential functions for the neural activation. The goal of decreasing the procedural risk factor in stereotactic surgery is accomplished by implementing the visual interface and a framework of artificial neural networks.>