Lakshmi Sugavaneswaran

Time-Frequency Analysis via Ramanujan Sums

Research in signal processing shows that a variety of transforms have been introduced to map the data from the original space into the feature space, in order to efficiently analyze a signal. These techniques differ in their basis functions, that is used for projecting the signal into a higher dimensional space. One of the widely used schemes for quasi-stationary and non-stationary signals is the time-frequency (TF) transforms, characterized by specific kernel functions. This work introduces a novel class of Ramanujan Fourier Transform (RFT) based TF transform functions, constituted by Ramanujan sums (RS) basis. The proposed special class of transforms offer high immunity to noise interference, since the computation is carried out only on co-resonant components, during analysis of signals. Further, we also provide a 2-D formulation of the RFT function. Experimental validation using synthetic examples, indicates that this technique shows potential for obtaining relatively sparse TF-equivalent representation and can be optimized for characterization of certain real-life signals.

Application of Empirical Mode Decomposition and Teager energy operator to EEG signals for mental task classification

This paper presents a novel method for mental task classification from EEG signals using Empirical Mode Decomposition and Teager energy operator techniques on EEG data. The efficacy of these techniques for non-stationary and non-linear data has already been demonstrated, which therefore lend themselves well to EEG signals, which are also non-stationary and non-linear in nature. The method described in this paper decomposed the EEG signals (6 EEG signals per task per subject, for a total of 5 tasks over multiple trials) into their constituent oscillatory modes, called intrinsic mode functions, and separated out the trend from the signal. Teager energy operator was used to calculate the average energy of both the detrended signal and the trend. The average energy was used to construct separate feature vectors with a small number of parameters for the detrended signal and the trend. Based on these parameters, one-versus-one classification of mental tasks was performed using Linear Discriminant Analysis. Using both feature vectors, an average correct classification rate of more than 85% was achieved, demonstrating the effectiveness of the method used. Furthermore, this method used all the intrinsic mode functions, as opposed to similar studies, demonstrating that the trend of the EEG signal also contains important discriminatory information.

Wavelet-based characterization of gait signal for neurological abnormalities

Studies conducted by the World Health Organization (WHO) indicate that over one billion suffer from neurological disorders worldwide, and lack of efficient diagnosis procedures affects their therapeutic interventions. Characterizing certain pathologies of motor control for facilitating their diagnosis can be useful in quantitatively monitoring disease progression and efficient treatment planning. As a suitable directive, we introduce a wavelet-based scheme for effective characterization of gait associated with certain neurological disorders. In addition, since the data were recorded from a dynamic process, this work also investigates the need for gait signal re-sampling prior to identification of signal markers in the presence of pathologies. To benefit automated discrimination of gait data, certain characteristic features are extracted from the wavelet-transformed signals. The performance of the proposed approach was evaluated using a database consisting of 15 Parkinsons disease (PD), 20 Huntingtons disease (HD), 13 Amyotrophic lateral sclerosis (ALS) and 16 healthy control subjects, and an average classification accuracy of 85% is achieved using an unbiased cross-validation strategy. The obtained results demonstrate the potential of the proposed methodology for computer-aided diagnosis and automatic characterization of certain neurological disorders.

Ambiguity domain-based identification of altered gait pattern in ALS disorder.

The onset of a neurological disorder, such as amyotrophic lateral sclerosis (ALS), is so subtle that the symptoms are often overlooked, thereby ruling out the option of early detection of the abnormality. In the case of ALS, over 75% of the affected individuals often experience awkwardness when using their limbs, which alters their gait, i.e. stride and swing intervals. The aim of this work is to suitably represent the non-stationary characteristics of gait (fluctuations in stride and swing intervals) in order to facilitate discrimination between normal and ALS subjects. We define a simple-yet-representative feature vector space by exploiting the ambiguity domain (AD) to achieve efficient classification between healthy and pathological gait stride interval. The stride-to-stride fluctuations and the swing intervals of 16 healthy control and 13 ALS-affected subjects were analyzed. Three features that are representative of the gait signal characteristics were extracted from the AD-space and are fed to linear discriminant analysis and neural network classifiers, respectively. Overall, maximum accuracies of 89.2% (LDA) and 100% (NN) were obtained in classifying the ALS gait.

Wavelet-based markers of ventricular fibrillation in optimizing human cardiac resuscitation

During cardiac resuscitation from ventricular fibrillation (VF) it would be helpful if we could monitor and predict the optimal state of the heart to be shocked into a perfusing rhythm. Real-time feedback of this state to the emergency medical staff (EMS) could improve the survival rate after resuscitation. In this paper, using real world out-of-the-hospital human VF data obtained during resuscitation by EMS personnel, we present the results of applying wavelet markers in predicting the shock outcomes. We also performed comparative analysis of 5 existing techniques (spectral and correlation based approaches) against the proposed wavelet markers. A database of 29 human VF tracings was extracted from the defibrillator recordings collected by the EMS personnel and was used to validate the waveform markers. The results obtained by the comparison of the wavelet based features with other spectral, and correlation-based features indicates that the proposed wavelet features perform well with an overall accuracy of 79.3% in predicting the shock outcomes and hence demonstrate potential to provide near real-time feedback to EMS personnel in optimizing resuscitation outcomes.

Quantification of fragmented QRS complex using intrinsic time-scale decomposition

Abstract The QRS complex recorded from the surface electrocardiogram (ECG) arises from electrical activation of the ventricular myocardium through the normal conduction system. The presence of a fragmented QRS (fQRS) complex reflects abnormal electrical activation and has been recently shown to identify patients with heart disease at risk of sudden cardiac death (SCD). The evaluation of fQRS is currently performed qualitatively by visual inspection which can be time consuming and subject to interpretation. Moreover, qualitative assessment of QRS for fragmentation may be insensitive to more subtle deflections in the QRS complex that may be equally prognostic. This study proposes an automated method to quantify QRS fractionation using intrinsic time-scale decomposition (ITD). Instantaneous morphology features are extracted from the decomposed QRS signal to index variations in its shapes. Our quantitative fQRS metric was found to be significantly greater in QRS complexes with fragmentation compared to normal QRS complexes derived from real-world ECGs in the Physikalisch-Technische Bundesanstalt (PTB) database. ROC analysis showed an area under the curve of 0.96 when fQRS was quantified from the precordial ECG leads, V1–V6. Thus, quantification of fQRS using the proposed ITD-based method can accurately identify fQRS. Our approach shows tremendous potential and could be investigated further for SCD risk assessment in patients with heart disease by improving the identification of fQRS that may or may not be visually apparent.

Advanced K-means clustering algorithm for large ECG data sets based on K-SVD approach

Wireless electrocardiography (ECG) systems are crucial in detecting and diagnosing heart disorders. Minimizing power consumption and sampling-rate should be the key aspects when designing wireless ECG systems. In order to achieve portability coupled with ultra-low power consumption and sampling-rate, clustering and classification algorithms play an important role in developing wireless ECG systems. Currently used algorithms do have their share of drawbacks: 1) clustering and classification cannot be done in real time; 2) implementing existing algorithms would lead to higher computational costs. These drawbacks motivated us in developing novel optimized clustering algorithm which could easily scan large ECG datasets for characteristic bio-markers. In this paper, we present an advanced K-means clustering algorithm based on K-Singular Value Decomposition (K-SVD) approach with a connection to Compressed Sensing (CS) theory, followed by sorting the data using a K-Nearest Neighbours (K-NN) classifier. The proposed algorithm outperforms existing algorithms by achieving a classification accuracy of 99.3%. This ability allows reducing 15% of Average Classification Error (ACE). The proposed algorithm also reduces the clustering energy consumption by increasing the classification performance.

Supervised retinal biometrics in different lighting conditions

Retinal image has been considered for number of health and biometrics applications. However, the reliability of these has not been investigated thoroughly. The variation observed in retina scans taken at different times is attributable to differences in illumination and positioning of the camera. It causes some missing bifurcations and crossovers from the retinal vessels. Exhaustive selection of optimal parameters is needed to construct the best similarity metrics equation to overcome the incomplete landmarks. In this paper, we extracted multiple features from the retina scans and employs supervised classification to overcome the shortcomings of the current techniques. The experimental results of 60 retina scans with different lightning conditions demonstrate the efficacy of this technique. The results were compared with the existing methods.

Exploiting the ambiguity domain for non-stationary biomedical signal classification

Research in time-frequency distributions (TFDs) is limited in terms of their use of the available spatial domains and in their target applications. Most of the work up till now has been concentrated mainly on the t-ƒ domain space. This work presents a detailed study about the ambiguity domain (AD), their resemblance in the t-ƒ space and the significance of using such a representation. Further, a novel approach for the analysis and classification between normal and pathological speech signals is also provided. The quantitative measures obtained show comparable performance scores with the existing schemes. Evidently, gait from 51-normal and 161-abnormal subjects were studied and classified in this analysis. Results obtained from the quantitative analysis illustrate comparable performance characteristics with some of the recent schemes and a maximum classification accuracy of 97.5% is obtained.

Discriminative time-frequency kernels for gait analysis for amyotrophic lateral sclerosis

Many stochastic systems show certain trends which in turn govern their underlying non-stationary time varying behavior. In order to facilitate efficient quantification of such signals, their analysis necessitates the use of robust tools for discerning between different classes of data. Research show that, use of time-frequency techniques offer intelligible representations for non-stationary signals, along with facilitating computation of instantaneous parameters. Further, in order to obtain efficient discrimination machine learning (ML) modules are often used alongside suitable representation techniques. In this work, we exploit the concepts of ML-kernel functions directly by incorporating them in the ambiguity time-frequency (TF) space, thereby obtaining a one-step discrimination between different non-stationary patterns. The proposed technique is evaluated for quantification applications for gait signal analysis. An overall classification accuracy of 93.1% is reported for the neurological gait database consisting of signals from 16-control and 13-amyotrophic lateral sclerosis (ALS) subjects. Results indicate that this scheme offers great potential in designing robust tools for time-varying signal analysis.