Various machine-learning classification techniques have been employed previously to classify brain states in healthy and disease populations using functional magnetic resonance imaging (fMRI). These methods generally use supervised classifiers that are sensitive to outliers and require labeling of training data to generate a predictive model. Density-based clustering, which overcomes these issues, is a popular unsupervised learning approach whose utility for high-dimensional neuroimaging data has not been previously evaluated. Its advantages include insensitivity to outliers and ability to work with unlabeled data. Unlike the popular k -means clustering, the number of clusters need not be specified. In this study, we compare the performance of two popular density-based clustering methods, DBSCAN and OPTICS, in accurately identifying individuals with three stages of cognitive impairment, including Alzheimer’s disease. We used static and dynamic functional connectivity features for clustering, which captures the strength and temporal variation of brain connectivity respectively. To assess the robustness of clustering to noise/outliers, we propose a novel method called recursive-clustering using additive-noise (R-CLAN). Results demonstrated that both clustering algorithms were effective, although OPTICS with dynamic connectivity features outperformed in terms of cluster purity (95.46%) and robustness to noise/outliers. This study demonstrates that density-based clustering can accurately and robustly identify diagnostic classes in an unsupervised way using brain connectivity.

In pulse diagnosis, the pulse signals obtained at wrist have been used for analysis of certain diseases in ancient systems of medicine in which the practitioner feels the pulse of the subject by placing his three fingers on the subject’s wrist at three distinct radial pulse point locations. The preliminary studies show that there are many conventional linear techniques applied to analyze the wrist pulse signals and less focus on non-linear techniques. Hence, the main aim of this research is to apply Recurrence Plot and Recurrence Quantification Analysis (RQA), a nonlinear technique to analyze the wrist pulse signals for distinguishing between diabetic and non-diabetic subjects. Wrist pulse signals from 32 subjects were recorded during the morning hours and were analyzed using RQA techniques. The results show significant differences in the RQA parameters of the wrist pulse signals as they are obtained from the recurrences occurring in the phase space plots of the wrist pulse signals. It was found that parameters like entropy, divergence and average diagonal line length showed significant variations for diabetic and nondiabetic subjects. Therefore, it can be concluded that RQA parameters can be used effectively to identify diabetic and nondiabetic subjects and thus may be applied on the wrist pulse signals for early detection of various diseases.

Music is one of the major activities that alters the emotional experience of a person. Musical processing in the brain is a complex process involving coordination between various areas of the brain. There are less number of studies that focus on analyzing brain responses due to music using modern signal processing techniques. This research aims to apply a nonlinear signal processing technique i.e. the Recurrence Quantification Analysis (RQA) technique to analyze the brain correlates of happy and sad music conditions while listening to happy and sad ragas of North Indian Classical Music (NICM). EEG signals from 20 different subjects are acquired while listening to excerpts of raga elaboration phases of NICM. Along with behavioural ratings, the signals were analyzed using the Recurrence Quantification Analysis technique. The results showed significant differences in the recurrence plot and recurrence parameters extracted from the frontal and fronto temporal regions in the right and left hemispheres of the brain. Therefore, from the results, it can be concluded that RQA parameters can detect emotional changes due to happy and sad music conditions.

Wrist pulse signals contain information about the status of health of a person and hence diagnosis using these signals is of great importance since long time. In this paper the efficacy of signal processing techniques in extracting useful information from wrist pulse signals has been demonstrated by using signals recorded before lunch and after lunch conditions. We have used correlation between probabilities of recurrence which is a recent nonlinear measure of synchronization, as well as Pearson's correlation coefficient, which is a traditionally used measure of synchronization. Statistical analysis using two-sample t-test is performed, which shows that there is significantly different synchronizations between signals acquired before lunch and those acquired after lunch. Moreover, upon performing RCE-SVM classification, the nonlinear synchronization measure showed a classification accuracy of 91.6% against the accuracy of 80.5% for correlation. These results show that the new nonlinear synchronization measure is effective in distinguishing the changes taking place after having lunch. This paper also demonstrates the ability of the wrist pulse signals in detecting changes occurring under two different conditions. This study is significant given that analysis of wrist pulse signals in case of food intake is less explored in literature and also that it has potential applications in healthcare.

Clustering is one of the most important methods for organizing database into groups. In this paper, Ordering Points To Identify Clustering Structure (OPTICS) algorithm has been used to perform clustering of functional Magnetic Resonance Imaging (fMRI) data. Dynamic functional connectivity features on fMRI data (ADNI database) obtained from subjects with early mild cognitive impairment (E-MCI), late mild cognitive impairment (L-MCI), Alzheimer's disease and healthy controls has been used for the study. On performing clustering, it has been observed that OPTICS is able to cluster the subjects into four inherent groups with a very high success rate. This result gives rise to applications in determining latent groups indicating various brain disorders.

One of the biggest single cause of mortality in the world is cardiovascular disease (CVD) and hence it is recognized that an early detection of its onset is vital for effective preventive therapies. Aortic stiffness has been shown to be an independent predictor of CVD, but its measurement is complex and time consuming. An alternative much simpler way of doing it is by looking for arterial properties such as arterial stiffness. This paper presents methods of estimating arterial parameters from measured ECG and PPG signals. PPG signal can be acquired easily by measuring the transmission of infra-red light through the finger tip. Two methods of obtaining arterial parameters have been used, one using simultaneously acquired ECG and PPG signals and another using PPG signal only. Pulse wave velocity, which is an indicator of arterial stiffness, has been measured using pulse wave transmit time derived from simultaneously recorded ECG and PPG signals. Arterial parameters like stiffness index, augmentation index etc have been derived from the PPG signals. These parameters have been obtained for subjects in two age groups and it is observed that the parameters show clear changes for the two age groups thus demonstrating their efficacy in detecting changes in the arteries with age. Power spectral analysis has been performed to show how frequency spectra are different for the two age groups. This study demonstrates the usefulness of ECG and PPG analysis in investigating the changes in the arterial characteristics of the vascular system with age; it has potential applications in healthcare.

Wrist pulse signal contains more important information about the health status of a person and pulse signal diagnosis has been employed in oriental medicine since very long time. In this paper we have used signal processing techniques to extract information from wrist pulse signals. For this purpose we have acquired radial artery pulse signals at wrist position noninvasively for different cases of interest. The wrist pulse waveforms have been analyzed using spatial features. Results have been obtained for the case of wrist pulse signals recorded for several subjects before exercise and after exercise. It is shown that the spatial features show statistically significant changes for the two cases and hence they are effective in distinguishing the changes taking place due to exercise. Support vector machine classifier is used to classify between the groups, and a high classification accuracy of 99.71% is achieved. Thus this paper demonstrates the utility of the spatial features in studying wrist pulse signals obtained under various recording conditions. The ability of the model to distinguish changes occurring under two different recording conditions can be potentially used for healthcare applications.

Wrist pulse signals contain important information about the health of a person and hence diagnosis based on pulse signals has assumed great importance. In this paper we demonstrate the efficacy of a two term Gaussian model to extract information from pulse signals. Results have been obtained by conducting experiments on several subjects to record wrist pulse signals for the cases of before exercise and after exercise. Parameters have been extracted from the recorded signals using the model and a paired t-test is performed, which shows that the parameters are significantly different between the two groups. Further, a recursive cluster elimination based support vector machine is used to perform classification between the groups. An average classification accuracy of 99.46% is obtained, along with top classifiers. It is thus shown that the parameters of the Gaussian model show changes across groups and hence the model is effective in distinguishing the changes taking place due to the two different recording conditions. The study has potential applications in healthcare.

This paper deals with processing the EEG signals obtained from 16 spatially arranged electrodes to measure coupling or synchrony between the frontal, parietal, occipital and temporal lobes of the cerebrum under the eyes open and eyes closed conditions. This synchrony was measured using magnitude squared coherence, Short Time Fourier Transform and wavelet based coherences. We found a pattern in the time-frequency coherence as we moved from the nasion to the inion of the subject's head. The coherence pattern obtained from the wavelet approach was found to be far more capable of picking up peaks in coherence with respect to frequency when compared to the regular Fourier based coherence. We detected high synchrony between frontal polar electrodes that is missing in coherence plots between other electrode pairs. The study has potential applications in healthcare.

The major difficulty in handling time series measurements is in identifying the system, whether it is purely deterministic, chaotic, or random. Proper identification of system characteristics and application of appropriate signal processing tools can lead to superior performance in signal analysis; especially when one handles real speech signals, these issues become even more important and crucial. Evidence for chaotic behavior with speech signals has been claimed and disputed. Over the past two decades, researchers have come out with efficient nonlinear dynamical tools applicable to time series. In this paper, different nonlinear dynamical tools like phase-space plot, running correlation dimension, and running Lyapunov exponent applied to speech signals are discussed which provide a convenient framework for speech signal analysis.

The problems encountered m fhe study of vibrating systems can be broad!, classified into tno main categories of analysis and synthesis. This p o w is concerned with a third category, viz., given a system and a prescrzbed set offorces acting on it, how can the spatial response of the system 6e altered in a desired manner by the application of additional (control) forces? In particular we consider the problem of controlling the energy of vibration of a driven string over aportion of its length by applyjog two control forces. Sfarting from graphical considerations m analyticai method has been deduced. The results show that a good controi ir possible. The effect of varying the point of application of a single force is then discrissed.

Whitespace identification is a crucial first-step in the implementation of cognitive radios, where the problem is to determine the communication footprint of active primary transmitters in a given geographical area. To do this, a number of sensors are deployed at known locations chosen uniformly at random within the given area. The sensors’ decisions regarding the presence or absence of a signal at their location is transmitted to a fusion center, which then combines the received information to construct the spatial spectral usage map. Under this model, several innovations are presented in this work to enable fast identification of the available whitespace. First, using the fact that a typical communication footprint is a sparse image, two novel compressed sensing based reconstruction methods are proposed to reduce the number of transmissions required from the sensors compared to a round-robin querying scheme. Second, a new method based on a combination of the K -means algorithm and a circular fitting technique is proposed for determining the number of primary transmitters. Third, a design procedure to determine the power thresholds for signal detection at sensors is discussed. The proposed schemes are experimentally compared with the round-robin scheme in terms of the average error in footprint identification relative to the area under consideration. Simulation results illustrate the improved performance of the proposed schemes relative to the round-robin scheme.

In this paper, we have proposed multiscale fractal dimension (MSFD) technique to characterize signals at multiple time scales. In this technique, multiple scales of the signal are obtained by segment averaging and the complexity of the resulting signals at those scales is quantified using multiresolution area-based fractal dimension measure. The technique is applied to intracranial EEG records and meditation HRV signals to detect change in states of physiological systems. We have considered two types of meditation techniques and pre-meditation state is used as control state against which MSFD parameters are group matched. The proposed MSFD technique has provided good performance and statistically significant results in discriminating epileptic seizures and meditation states from corresponding controls. The technique can be used in diverse applications of signal processing.

Electroencephalogram (EEG) recordings are often contaminated with ocular and muscle artifacts. In this paper, the canonical correlation analysis (CCA) is used as blind source separation (BSS) technique (BSS-CCA) to decompose the artifact contaminated EEG into component signals. We combine the BSSCCA technique with wavelet filtering approach for minimizing both ocular and muscle artifacts simultaneously, and refer the proposed method as wavelet enhanced BSS-CCA. In this approach, after careful visual inspection, the muscle artifact components are discarded and ocular artifact components are subjected to wavelet filtering to retain high frequency cerebral information, and then clean EEG is reconstructed. The performance of the proposed wavelet enhanced BSS-CCA method is tested on real EEG recordings contaminated with ocular and muscle artifacts, for which power spectral density is used as a quantitative measure. Our results suggest that the proposed hybrid approach minimizes ocular and muscle artifacts effectively, minimally affecting underlying cerebral activity in EEG recordings. Keywords—Blind source separation, Canonical correlation analysis, Electroencephalogram, Muscle artifact, Ocular artifact, Power spectrum, Wavelet threshold.

This paper considers the problem of identifying the footprints of communication of multiple transmitters in a given geographical area. To do this, a number of sensors are deployed at arbitrary but known locations in the area, and their individual decisions regarding the presence or absence of the transmitters' signal are combined at a fusion center to reconstruct the spatial spectral usage map. One straightforward scheme to construct this map is to query each of the sensors and cluster the sensors that detect the primary's signal. However, using the fact that a typical transmitter footprint map is a sparse image, two novel compressive sensing based schemes are proposed, which require significantly fewer number of transmissions compared to the querying scheme. A key feature of the proposed schemes is that the measurement matrix is constructed from a pseudo-random binary phase shift applied to the decision of each sensor prior to transmission. The measurement matrix is thus a binary ensemble which satisfies the restricted isometry property. The number of measurements needed for accurate footprint reconstruction is determined using compressive sampling theory. The three schemes are compared through simulations in terms of a performance measure that quantifies the accuracy of the reconstructed spatial spectral usage map. It is found that the proposed sparse reconstruction technique-based schemes significantly outperform the round-robin scheme.

Many recent electrophysiological studies have revealed the importance of investigating meditation state in order to achieve an increased understanding of autonomous control of cardiovascular functions. In this paper, we characterize heart rate variability (HRV) time series acquired during meditation using nonlinear dynamical parameters. We have computed minimum embedding dimension (MED), correlation dimension (CD), largest Lyapunov exponent (LLE), and nonlinearity scores (NLS) from HRV time series of eight Chi and four Kundalini meditation practitioners. The pre-meditation state has been used as a baseline (control) state to compare the estimated parameters. The chaotic nature of HRV during both pre-meditation and meditation is confirmed by MED. The meditation state showed a significant decrease in the value of CD and increase in the value of LLE of HRV, in comparison with premeditation state, indicating a less complex and less predictable nature of HRV. In addition, it was shown that the HRV of meditation state is having highest NLS than pre-meditation state. The study indicated highly nonlinear dynamic nature of cardiac states as revealed by HRV during meditation state, rather considering it as a quiescent state. Keywords—Correlation dimension, Embedding dimension, Heart rate variability, Largest Lyapunov exponent, Meditation, Nonlinearity score.

Fractal Dimensions (FD) are one of the popular measures used for characterizing signals. They have been used as complexity measures of signals in various fields including speech and biomedical applications. However, proper interpretation of such analyses has not been thoroughly addressed. In this paper, we study the effect of various signal properties on FD and interpret results in terms of classical signal processing concepts such as amplitude, frequency, number of harmonics, noise power and signal bandwidth. We have used Higuchi's method for estimating FDs. This study may help in gaining a better understanding of the FD complexity measure itself, and for interpreting changing structural complexity of signals in terms of FD. Our results indicate that FD is a useful measure in quantifying structural changes in signal properties.

In this paper, we have developed a method to compute fractal dimension (FD) of discrete time signals, in the time domain, by modifying the box-counting method. The size of the box is dependent on the sampling frequency of the signal. The number of boxes required to completely cover the signal are obtained at multiple time resolutions. The time resolutions are made coarse by decimating the signal. The loglog plot of total number of boxes required to cover the curve versus size of the box used appears to be a straight line, whose slope is taken as an estimate of FD of the signal. The results are provided to demonstrate the performance of the proposed method using parametric fractal signals. The estimation accuracy of the method is compared with that of Katz, Sevcik, and Higuchi methods. In ddition, some properties of the FD are discussed.