Palaniappan Ramaswamy | University of Kent (original) (raw)

Uploads

Papers by Palaniappan Ramaswamy

Research paper thumbnail of Neural network classification of homomorphic segmented heart sounds

Applied Soft Computing, 2007

Research paper thumbnail of Single-Trial Visual Evoked Potential Extraction Using Partial Least-Squares-Based Approach

IEEE Journal of Biomedical and Health Informatics, 2016

Research paper thumbnail of Fuzzy ATRMAP classification of mental tasks using segmented and overlapped EEG signals

Research paper thumbnail of Multi-parameter detection of ectopic heart beats

IEEE International Workshop on Biomedical Circuits and Systems, 2004.

... MULTI-PARAMETER DETECTION OF ECTOPIC HEART BEATS Ramaswamy Palaniappan, Cuta Navin Guptu, Cha... more ... MULTI-PARAMETER DETECTION OF ECTOPIC HEART BEATS Ramaswamy Palaniappan, Cuta Navin Guptu, Chan Kap Luk, Shankur M. Krishnan Biomedical Engineering Research Centre, Nanyang Technological University, Singapore ABSTRACT ...

Research paper thumbnail of Genetic algorithm to select features for fuzzy ARTMAP classification of evoked EEG

Asia-Pacific Conference on Circuits and Systems

ABSTRACT

Research paper thumbnail of Identifying individuals using ECG beats

2004 International Conference on Signal Processing & Communications (SPCOM), 2004

... The results gave classification performance up tu 97.6%. This indicates that ECG has the pote... more ... The results gave classification performance up tu 97.6%. This indicates that ECG has the potential to be used as a biometric tool. 1. INTRODUCTION The most common method of identifying individuals is through the use of fingerprints (thumbprints) [I, 21. ...

Research paper thumbnail of Detection of ectopic heartbeats using ECG and blood pressure signals

2004 International Conference on Signal Processing & Communications (SPCOM), 2004

ABSTRACT

Research paper thumbnail of Improving simplified fuzzy ARTMAP performance using genetic algorithm for brain fingerprint classification

Proceedings - 2006 14th International Conference on Advanced Computing and Communications, ADCOM 2006, 2006

Fuzzy ARTMAP (FA) [1] is an incremental neural network classifier, which has found use in numerou... more Fuzzy ARTMAP (FA) [1] is an incremental neural network classifier, which has found use in numerous pattern recognition problems [2, 3]. Simplified Fuzzy ARTMAP (SFA) is a simpler version of FA, which is faster and performs equally well as FA [4, 5]. However, both the generic FA ...

Research paper thumbnail of Effective Normalisation of Selective Eigen Rate Method to Separate Principal Components of Vep and Eeg

Here we give proof to the best suitable normalization method for the Selective Eig en Rate (SER) ... more Here we give proof to the best suitable normalization method for the Selective Eig en Rate (SER) a novel technique, that is used in selecting only the higher rate of principal components (PCs) for using them in Principal Component Analysis (PCA) while separating Visual Evoked Potential (VEP) from electroencephalogram (EEG) signals, to enable single trial analysis. SER technique is designed and implemented to overcome heavy electroencephalogram (EEG) contamination in VEP signals. Normalisation of the eigen values which are obtained as a result of PCA is an important part for PC selection process in SER technique. In order to derive the maximum signal to noise ratio (SNR) from artificial VEP signals contaminated by EEG, three distinct normalisation methods were constructed and tested here. The best methods of normalisation suitable for the SER method is found and tested with added factors of noise in multiples of 2, 5 and 10 times. Assessment on the performance of this effective norma...

Research paper thumbnail of Mental task discrimination using EEG signals with genetically trained fuzzy ARTMAP

We show that it is possible to discriminate accurately between human mental tasks based on electr... more We show that it is possible to discriminate accurately between human mental tasks based on electroencephalogram (EEG) signals only. We study two different mental tasks performed by four subjects across two different sessions. The EEG power spectral densities (PSD) from 0 to 30 Hz are extracted using Wiener Khintchine theorem with Parzen window smoothing and are trained with a Fuzzy ARTMAP classifier (FA) [1] instead of frequency band power values commonly used. FA classification performance varies for different order of pattern feeding. To solve this problem, we propose a novel method of fusing genetic algorithms (GA) instead of voting strategy proposed in [1] and Table 1 shows the improvement of our method over the latter.

Research paper thumbnail of Time-series analysis of EEG signals

Research paper thumbnail of VEP spectral bands for detecting alcoholics

Research paper thumbnail of Autoregressive order selection criteria: a case study for EEG signals

Autoregressive (AR) models have a broad spectrum of applications ranging from identification; pre... more Autoregressive (AR) models have a broad spectrum of applications ranging from identification; prediction and control of dynamical systems and digital spectral analysis using these models have proven to be superior to classical Fourier transform techniques. However, there is a parameter that must be selected to utilize an AR technique properly i.e. the model order. Therefore, methods that will determine the appropriate model order must be used. In this paper, we perform a case study of the currently available statistical methods like Akaike Information Criterion, Final Prediction Error, Residual Variance, Minimum Description Length, Criterion Autoregressive Transfer and HannanQuinn. These methods depend on the statistical properties of the data, which selects the lowest order that is optimal to represent the signal. We use a Fuzzy ARTMAP (FA) neural network to study the performance of the different statistical criteria to select the appropriate AR model order for EEG signal representations. We perform this by training the FA to classify different mental tasks using the spectrum obtained from AR analysis with the selected model order. In addition to the other statistical methods, we have also studied the performance of a fixed 6 th order AR model.

Research paper thumbnail of Neural network classification of brain waves using asymmetry ratio

Research paper thumbnail of Gamma band analysis of VEP to study the electrophysiological differences in alcoholics

Introduction In this paper, we analyse Visual Evoked Potential (VEP) in the gamma band range of 3... more Introduction In this paper, we analyse Visual Evoked Potential (VEP) in the gamma band range of 30-50 Hz to study the electrophysiological differences between alcoholics and non-alcoholics. Gamma band spectrum is used specifically in the analysis because it has been shown to be closely related to higher brain functions like memory and object recognition [1]. However, the use of gamma band spectrum to analyse electrophysiological differences in alcoholics is novel. The VEP signals are extracted from 64 electrodes while the subjects are seeing 2 visual stimuli (presented with an interval in-between) from the Snodgrass and Vanderwart picture set. The experimental paradigm is designed to evoke visual short-term memory and object recognition abilities. Twenty subjects participated in the experimental study consisting of 10 alcoholics and 10 non-alcoholics. Forward and reverse Butterworth digital filter is used to extract VEP signals in gamma band spectral range. Parseval’s theorem is used obtain the equivalent gamma band spectral power. The results using t-Test analysis indicate that alcoholics give lower gamma band spectral power as compared to nonalcoholics in certain channels located in the central, occipital and parietal regions. This shows that some alterations to the brain processes that involve visual shortterm memory and object recognition are caused by longterm use of alcohol. The nature of these alterations is still traceable after a period of time, which is indicated by the fact that the studied alcoholics had been abstinent for a period of more than a month.

Research paper thumbnail of Analysis of P3 visual stimulus EEG

In this paper, visual stimulus EEG (VSE) signals are extracted during a modified delayed matching... more In this paper, visual stimulus EEG (VSE) signals are extracted during a modified delayed matching-to-sample paradigm. These VSE signals are used to investigate the differences in object recognition and decision-making process between non-amnesic alcoholic and nonalcoholic subjects. P3 responses are used in the investigation since they are widely associated with object recognition and decision-making ability. Our experimental results indicate that the P3 responses show differences between alcoholics and non-alcoholics. Specifically, the results indicate that alcoholics exhibits lower P3 amplitude and slower P3 response as compared to non-alcoholics.

Research paper thumbnail of Classification of alcoholics using gamma band spectral power ratio extracted from single-trial and de-noised VEP signals

Introduction In this paper, we classify alcoholics and non-alcoholics using gamma band spectral p... more Introduction In this paper, we classify alcoholics and non-alcoholics using gamma band spectral power ratio extracted from single trials of de-noised Visual Evoked Potential (VEP) signals. Simplified Fuzzy ARTMAP (SFA) neural network is used in the classification. Principal Component Analysis (PCA) is used to reduce the effects of noise. Single trial analysis is possible in our case because gamma band spectrum is above 30 Hz. As such the averaging procedure that is necessary to remove background electroencephalogram (EEG), which is normally below 30 Hz, is circumvented.

Research paper thumbnail of Evolutionary fuzzy ARTMAP for optimal classification of evoked EEG

Research paper thumbnail of Autoregressive modelling of pseudo-periodic VEP signals for classification of alcoholics

Research paper thumbnail of Extracting single trial evoked potential signals using spectral power ratio principal components

Spectral Power Ratio (SPR) is a novel technique proposed that selects only the specific principal... more Spectral Power Ratio (SPR) is a novel technique proposed that selects only the specific principal components (PCs) in Principal Component Analysis (PCA) for single trial extraction of evoked potential (EP) signals. SPR technique is designed and implemented to overcome heavy electroencephalogram (EEG) contamination in the EP signals. By using SPR technique, increased signal to noise ratio (SNR) is obtained when applied on artificial Visual EP (VEP) signals contaminated by EEG. The EEG is added with factors of 2,4,6,8, and 10 times to test SPR’s suitability to be used with real VEP. Assessment on these signals shows that application of SPR on contaminated signals outperformed the existing Kaiser (KSR) and Residual Power (RP) methods to select the PCs. The results proved SPR’s consistent performance with an average SNR of 1.204 dB while RP and KSR gave –0.24 dB and KSR –0.36 dB, respectively where the original SNR was –7.08. Usefulness of SPR is confirmed using real VEP signals to analyse P3 amplitude and latency responses to matching and non-matching visual stimuli. SPR extracted P3 parameters resulted in faster and higher responses (p<0.05) for the matched stimuli, which confirms existing neuroscience knowledge while KSR and RP methods fail to indicate significant differences.

Research paper thumbnail of Neural network classification of homomorphic segmented heart sounds

Applied Soft Computing, 2007

Research paper thumbnail of Single-Trial Visual Evoked Potential Extraction Using Partial Least-Squares-Based Approach

IEEE Journal of Biomedical and Health Informatics, 2016

Research paper thumbnail of Fuzzy ATRMAP classification of mental tasks using segmented and overlapped EEG signals

Research paper thumbnail of Multi-parameter detection of ectopic heart beats

IEEE International Workshop on Biomedical Circuits and Systems, 2004.

... MULTI-PARAMETER DETECTION OF ECTOPIC HEART BEATS Ramaswamy Palaniappan, Cuta Navin Guptu, Cha... more ... MULTI-PARAMETER DETECTION OF ECTOPIC HEART BEATS Ramaswamy Palaniappan, Cuta Navin Guptu, Chan Kap Luk, Shankur M. Krishnan Biomedical Engineering Research Centre, Nanyang Technological University, Singapore ABSTRACT ...

Research paper thumbnail of Genetic algorithm to select features for fuzzy ARTMAP classification of evoked EEG

Asia-Pacific Conference on Circuits and Systems

ABSTRACT

Research paper thumbnail of Identifying individuals using ECG beats

2004 International Conference on Signal Processing & Communications (SPCOM), 2004

... The results gave classification performance up tu 97.6%. This indicates that ECG has the pote... more ... The results gave classification performance up tu 97.6%. This indicates that ECG has the potential to be used as a biometric tool. 1. INTRODUCTION The most common method of identifying individuals is through the use of fingerprints (thumbprints) [I, 21. ...

Research paper thumbnail of Detection of ectopic heartbeats using ECG and blood pressure signals

2004 International Conference on Signal Processing & Communications (SPCOM), 2004

ABSTRACT

Research paper thumbnail of Improving simplified fuzzy ARTMAP performance using genetic algorithm for brain fingerprint classification

Proceedings - 2006 14th International Conference on Advanced Computing and Communications, ADCOM 2006, 2006

Fuzzy ARTMAP (FA) [1] is an incremental neural network classifier, which has found use in numerou... more Fuzzy ARTMAP (FA) [1] is an incremental neural network classifier, which has found use in numerous pattern recognition problems [2, 3]. Simplified Fuzzy ARTMAP (SFA) is a simpler version of FA, which is faster and performs equally well as FA [4, 5]. However, both the generic FA ...

Research paper thumbnail of Effective Normalisation of Selective Eigen Rate Method to Separate Principal Components of Vep and Eeg

Here we give proof to the best suitable normalization method for the Selective Eig en Rate (SER) ... more Here we give proof to the best suitable normalization method for the Selective Eig en Rate (SER) a novel technique, that is used in selecting only the higher rate of principal components (PCs) for using them in Principal Component Analysis (PCA) while separating Visual Evoked Potential (VEP) from electroencephalogram (EEG) signals, to enable single trial analysis. SER technique is designed and implemented to overcome heavy electroencephalogram (EEG) contamination in VEP signals. Normalisation of the eigen values which are obtained as a result of PCA is an important part for PC selection process in SER technique. In order to derive the maximum signal to noise ratio (SNR) from artificial VEP signals contaminated by EEG, three distinct normalisation methods were constructed and tested here. The best methods of normalisation suitable for the SER method is found and tested with added factors of noise in multiples of 2, 5 and 10 times. Assessment on the performance of this effective norma...

Research paper thumbnail of Mental task discrimination using EEG signals with genetically trained fuzzy ARTMAP

We show that it is possible to discriminate accurately between human mental tasks based on electr... more We show that it is possible to discriminate accurately between human mental tasks based on electroencephalogram (EEG) signals only. We study two different mental tasks performed by four subjects across two different sessions. The EEG power spectral densities (PSD) from 0 to 30 Hz are extracted using Wiener Khintchine theorem with Parzen window smoothing and are trained with a Fuzzy ARTMAP classifier (FA) [1] instead of frequency band power values commonly used. FA classification performance varies for different order of pattern feeding. To solve this problem, we propose a novel method of fusing genetic algorithms (GA) instead of voting strategy proposed in [1] and Table 1 shows the improvement of our method over the latter.

Research paper thumbnail of Time-series analysis of EEG signals

Research paper thumbnail of VEP spectral bands for detecting alcoholics

Research paper thumbnail of Autoregressive order selection criteria: a case study for EEG signals

Autoregressive (AR) models have a broad spectrum of applications ranging from identification; pre... more Autoregressive (AR) models have a broad spectrum of applications ranging from identification; prediction and control of dynamical systems and digital spectral analysis using these models have proven to be superior to classical Fourier transform techniques. However, there is a parameter that must be selected to utilize an AR technique properly i.e. the model order. Therefore, methods that will determine the appropriate model order must be used. In this paper, we perform a case study of the currently available statistical methods like Akaike Information Criterion, Final Prediction Error, Residual Variance, Minimum Description Length, Criterion Autoregressive Transfer and HannanQuinn. These methods depend on the statistical properties of the data, which selects the lowest order that is optimal to represent the signal. We use a Fuzzy ARTMAP (FA) neural network to study the performance of the different statistical criteria to select the appropriate AR model order for EEG signal representations. We perform this by training the FA to classify different mental tasks using the spectrum obtained from AR analysis with the selected model order. In addition to the other statistical methods, we have also studied the performance of a fixed 6 th order AR model.

Research paper thumbnail of Neural network classification of brain waves using asymmetry ratio

Research paper thumbnail of Gamma band analysis of VEP to study the electrophysiological differences in alcoholics

Introduction In this paper, we analyse Visual Evoked Potential (VEP) in the gamma band range of 3... more Introduction In this paper, we analyse Visual Evoked Potential (VEP) in the gamma band range of 30-50 Hz to study the electrophysiological differences between alcoholics and non-alcoholics. Gamma band spectrum is used specifically in the analysis because it has been shown to be closely related to higher brain functions like memory and object recognition [1]. However, the use of gamma band spectrum to analyse electrophysiological differences in alcoholics is novel. The VEP signals are extracted from 64 electrodes while the subjects are seeing 2 visual stimuli (presented with an interval in-between) from the Snodgrass and Vanderwart picture set. The experimental paradigm is designed to evoke visual short-term memory and object recognition abilities. Twenty subjects participated in the experimental study consisting of 10 alcoholics and 10 non-alcoholics. Forward and reverse Butterworth digital filter is used to extract VEP signals in gamma band spectral range. Parseval’s theorem is used obtain the equivalent gamma band spectral power. The results using t-Test analysis indicate that alcoholics give lower gamma band spectral power as compared to nonalcoholics in certain channels located in the central, occipital and parietal regions. This shows that some alterations to the brain processes that involve visual shortterm memory and object recognition are caused by longterm use of alcohol. The nature of these alterations is still traceable after a period of time, which is indicated by the fact that the studied alcoholics had been abstinent for a period of more than a month.

Research paper thumbnail of Analysis of P3 visual stimulus EEG

In this paper, visual stimulus EEG (VSE) signals are extracted during a modified delayed matching... more In this paper, visual stimulus EEG (VSE) signals are extracted during a modified delayed matching-to-sample paradigm. These VSE signals are used to investigate the differences in object recognition and decision-making process between non-amnesic alcoholic and nonalcoholic subjects. P3 responses are used in the investigation since they are widely associated with object recognition and decision-making ability. Our experimental results indicate that the P3 responses show differences between alcoholics and non-alcoholics. Specifically, the results indicate that alcoholics exhibits lower P3 amplitude and slower P3 response as compared to non-alcoholics.

Research paper thumbnail of Classification of alcoholics using gamma band spectral power ratio extracted from single-trial and de-noised VEP signals

Introduction In this paper, we classify alcoholics and non-alcoholics using gamma band spectral p... more Introduction In this paper, we classify alcoholics and non-alcoholics using gamma band spectral power ratio extracted from single trials of de-noised Visual Evoked Potential (VEP) signals. Simplified Fuzzy ARTMAP (SFA) neural network is used in the classification. Principal Component Analysis (PCA) is used to reduce the effects of noise. Single trial analysis is possible in our case because gamma band spectrum is above 30 Hz. As such the averaging procedure that is necessary to remove background electroencephalogram (EEG), which is normally below 30 Hz, is circumvented.

Research paper thumbnail of Evolutionary fuzzy ARTMAP for optimal classification of evoked EEG

Research paper thumbnail of Autoregressive modelling of pseudo-periodic VEP signals for classification of alcoholics

Research paper thumbnail of Extracting single trial evoked potential signals using spectral power ratio principal components

Spectral Power Ratio (SPR) is a novel technique proposed that selects only the specific principal... more Spectral Power Ratio (SPR) is a novel technique proposed that selects only the specific principal components (PCs) in Principal Component Analysis (PCA) for single trial extraction of evoked potential (EP) signals. SPR technique is designed and implemented to overcome heavy electroencephalogram (EEG) contamination in the EP signals. By using SPR technique, increased signal to noise ratio (SNR) is obtained when applied on artificial Visual EP (VEP) signals contaminated by EEG. The EEG is added with factors of 2,4,6,8, and 10 times to test SPR’s suitability to be used with real VEP. Assessment on these signals shows that application of SPR on contaminated signals outperformed the existing Kaiser (KSR) and Residual Power (RP) methods to select the PCs. The results proved SPR’s consistent performance with an average SNR of 1.204 dB while RP and KSR gave –0.24 dB and KSR –0.36 dB, respectively where the original SNR was –7.08. Usefulness of SPR is confirmed using real VEP signals to analyse P3 amplitude and latency responses to matching and non-matching visual stimuli. SPR extracted P3 parameters resulted in faster and higher responses (p<0.05) for the matched stimuli, which confirms existing neuroscience knowledge while KSR and RP methods fail to indicate significant differences.