Showing papers by "Samarendra Dandapat published in 2015"

Multiscale Energy and Eigenspace Approach to Detection and Localization of Myocardial Infarction

Abstract: In this paper, a novel technique on a multiscale energy and eigenspace (MEES) approach is proposed for the detection and localization of myocardial infarction (MI) from multilead electrocardiogram (ECG). Wavelet decomposition of multilead ECG signals grossly segments the clinical components at different subbands. In MI, pathological characteristics such as hypercute T-wave, inversion of T-wave, changes in ST elevation, or pathological Q-wave are seen in ECG signals. This pathological information alters the covariance structures of multiscale multivariate matrices at different scales and the corresponding eigenvalues. The clinically relevant components can be captured by eigenvalues. In this study, multiscale wavelet energies and eigenvalues of multiscale covariance matrices are used as diagnostic features. Support vector machines (SVMs) with both linear and radial basis function (RBF) kernel and K-nearest neighbor are used as classifiers. Datasets, which include healthy control, and various types of MI, such as anterior, anteriolateral, anterioseptal, inferior, inferiolateral, and inferioposterio-lateral, from the PTB diagnostic ECG database are used for evaluation. The results show that the proposed technique can successfully detect the MI pathologies. The MEES approach also helps localize different types of MIs. For MI detection, the accuracy, the sensitivity, and the specificity values are 96%, 93%, and 99% respectively. The localization accuracy is 99.58%, using a multiclass SVM classifier with RBF kernel.

A Gaussian Scale Space Approach For Exudates Detection, Classification And Severity Prediction.

Abstract: In the context of Computer Aided Diagnosis system for diabetic retinopathy, we present a novel method for detection of exudates and their classification for disease severity prediction. The method is based on Gaussian scale space based interest map and mathematical morphology. It makes use of support vector machine for classification and location information of the optic disc and the macula region for severity prediction. It can efficiently handle luminance variation and it is suitable for varied sized exudates. The method has been probed in publicly available DIARETDB1V2 and e-ophthaEX databases. For exudate detection the proposed method achieved a sensitivity of 96.54% and prediction of 98.35% in DIARETDB1V2 database.

A novel breathiness feature for analysis and classification of speech under stress

Abstract: This work explores the effect of breathiness component on speech under stress. The breathiness component in a speech signal can be estimated using different features such as period perturbation quotient (PPQ), amplitude perturbation quotient (APQ), harmonic to noise ratio (HNR), glottal to noise excitation ratio (GNER), harmonic energy (HE), harmonic energy of residue (HER) and harmonic to signal ratio (HSR). Statistical analysis of these features shows that they have different mean and variance values for speech under stress. The performance of breathiness features is evaluated using Hidden Markov Model (HMM) for classification of speech under stress. The results show that the breathiness features successfully characterize the speech under stress. The performance of breathiness features is compared with the MFCC feature. Finally, a speech under stress classification method is proposed with the combination of breathiness and MFCC features. In terms of classification rates, the proposed combined feature outperforms the MFCC feature.

Quantification of Diagnostic Information from Electrocardiogram Signal: A Review

Abstract: Electrocardiogram (ECG) contains the information about the contraction and relaxation of heart chambers. This diagnostic information will change due to various cardiovascular diseases. This information is used by a cardiologist for accurate detection of various life-threatening cardiac disorders. ECG signals are subjected to number of processing, for computer aided detection and localization of cardiovascular diseases. These processing schemes are categorized as filtering, synthesis, compression and transmission. Quantifying diagnostic information from an ECG signal in an efficient way, is always a challenging task in the area of signal processing. This paper presents a review on state-of-art diagnostic information extraction approaches and their applications in various ECG signal processing schemes such as quality assessment and cardiac disease detection. Then, a new diagnostic measure for multilead ECG (MECG) is proposed. The proposed diagnostic measure (MSD) is defined as the difference between multivariate sample entropy values for original and processed MECG signals. The MSD measure is evaluated over MECG compression framework. Experiments are conducted over both normal and pathological MECG from PTB database. The results demonstrate that the proposed MSD measure is effective in quantifying diagnostic information in MECG. The MSD measure is also compare with other measures such as WEDD, PRD and RMSE.

Exploiting multi-lead electrocardiogram correlations using robust third-order tensor decomposition.

Abstract: In this Letter, a robust third-order tensor decomposition of multi-lead electrocardiogram (MECG) comprising of 12-leads is proposed to reduce the dimension of the storage data. An order-3 tensor structure is employed to represent the MECG data by rearranging the MECG information in three dimensions. The three-dimensions of the formed tensor represent the number of leads, beats and samples of some fixed ECG duration. Dimension reduction of such an arrangement exploits correlations present among the successive beats (intra-beat and inter-beat) and across the leads (inter-lead). The higher-order singular value decomposition is used to decompose the tensor data. In addition, multiscale analysis has been added for effective care of ECG information. It grossly segments the ECG characteristic waves (P-wave, QRS-complex, ST-segment and T-wave etc.) into different sub-bands. In the meantime, it separates high-frequency noise components into lower-order sub-bands which helps in removing noise from the original data. For evaluation purposes, we have used the publicly available PTB diagnostic database. The proposed method outperforms the existing algorithms where compression ratio is under 10 for MECG data. Results show that the original MECG data volume can be reduced by more than 45 times with acceptable diagnostic distortion level.

Enhancement of the spatial resolution of ECG using multi-scale Linear Regression

Abstract: Spatial resolution of ECG can be increased using the information available from a subset of standard 12-lead ECG. This is usually achieved by learning a model between the standard 12-lead and its reduced lead subset. Since ECG signal contains significant amount of diagnostic information, it is important to learn a model which preserves this information. In this work, a patient specific model is proposed which utilizes the inter lead correlation in the transformed domain. The model is learned over Wavelet domain using Linear Regression. Performance of the model is evaluated using standard distortion measures such as correlation coefficient and root mean square error along with wavelet energy based diagnostic distortion. An analysis is also performed over the derived signal to quantify the loss of diagnostic information. The results show that the proposed model performs better in preserving diagnostic information in comparison to the existing linear models.

Linear transformation on speech subspace for analysis of speech under stress condition

Abstract: In this work, a novel approach of linear transformation on speech subspace is used to preserve the properties of speech signal under stress condition. It is assumed that, there is another subspace called as speech subspace which exist and contains the properties of speech signal under neutral and stress conditions. Therefore, speech component of stress speech is determined by linear transformation on speech subspace. The dimension of speech subspace is taken to be comparatively higher than original length of feature vector of training database to capture the variations in properties of speech signals more appropriately under stress condition. The linear transformation matrix is estimated using the information of HMM which is used to model the training database (neutral speech). The HMM information is used in terms of supervector. All the experiments in this work are done by parametrizing neutral and stress speech as nonlinear (TEO-CB-Auto-Env) feature. Experimentally it is observed that, a linear relationship exist between stress speech subspace and speech subspace. After linear transformation on speech subspace, speech recognizer outperforms by 7.57 % (62.14 % to 69.71%) under angry stress condition.

Automatic detection of fovea using property of vessel free region

Abstract: Accumulation of blood and its constituents over fovea of the retina lead to irreversible vision degradation in Diabetic Retinopathy (DR). Thus, fovea location contains very vital information in automated analysis. In this study, we have developed a simple approach for identification of fovea location. The main advantage of the method is that it does not require prior knowledge of the spatial relationship of optic disc location. The algorithm first searches for the fovea region considering the information that fovea is devoid of blood vessels. Later, dark intensity property of fovea is utilized for its detection from the region of interest. The method requires detection of blood vessel network and then search for vessel free region. Various morphological image processing tools are explored in different color planes for the successful execution of the method. The algorithm is tested on 759 images of DRIVE, DIARETDB0, DIARETDB1, LOCAL, MESSIDOR and HRF databases containing both normal and pathological cases of DR, with efficiency of detection obtained at 100%, 96.85%, 97.67%, 98.46% 96.25% and 100% respectively. The overall accuracy is 98.21%.

Cardiac arrhythmia classification from multilead ECG using multiscale non-linear analysis

Abstract: In this paper, a novel technique Is proposed for detecting cardiac arrhythmias using signals obtained from a multi-lead electrocardiogram (ECG). The method employs the use of two non-linear features namely detrended fluctuation analysis and sample entropy. The features are calculated on signals obtained after performing discrete wavelet transform on the incoming raw ECG data and selecting the diagnostically relevant sub-bands. The DFC and SE features of the sub-band signal are computed and the performance of these features is evaluated using multilayer perceptron (MLP), radial basis function neural network (RBFNN) and probabilistic neural network (PNN) classifiers. The experimental result shows that, the combination of DFC and SE features along-with MLP classifier has a high accuracy value of 98.76%.

Compressed sensing framework of data reduction at multiscale level for eigenspace multichannel ECG signals

Abstract: Multichannel elctrocardiogram (MECG) signals are correlated both in spatial domain as well as in temporal domain and this correlation becomes even higher at multiscale levels. This work presents a MECG compression method in order to exploit the inherent inter-channel correlation more efficiently, using a multiscale compressive sensing (MSCS) based approach. Principal component analysis (PCA) is used to decorrelate the subband signals from different channels at each wavelet scale and then the significant eigenspace signals from higher frequency subbands are undergone through multiscale compressed sensing (CS). Since CS is well known for its effective representation of high dimensional sparse signals in terms of few random projections, here it confines the noise dominated high frequency clinical information of MECG signals to few compressed measurements which readily reduces the data size at the encoder side. Eigenspace is taken as the sparsifying basis for high frequency subband ECG signals. The proposed encoding strategy is implemented using a uniform scalar quantizer and a entropy encoder. Sparse signal recovery is done using a greedy sparse recovery algorithm called orthogonal matching pursuit (OMP). Performance evaluation of the coder is mainly carried out in terms of compression ratio (CR), root mean square difference (PRD), and wavelet energy based diagnostic distortion (WEDD). Simulation results give the lowest PRD value, 4.72% and WEDD value 3.28% at CR=10.84, for lead aVF for CSE multi-lead measurement library database.

DCT-Based Linear Regression Approach for 12-Lead ECG Synthesis

Abstract: Synthesis of standard 12-lead electrocardiogram from reduced lead set without losing significant diagnostic information is a major challenge. In this work, we propose a patient specific method for synthesizing 12-lead electrocardiogram from reduced lead set by applying linear regression over the DCT domain. The proposed method is evaluated by standard distortion measures such as correlation coefficient, root mean square error, and wavelet energy-based diagnostic distortion. The results shows improvement from the existing systems without loss of significant diagnostic information.

Detection of cardiac ailments from multilead ECG using diagnostic eigen error features

Abstract: Accurate detection of life-threatening cardiac ailments is one of the important task in monitoring patient's health. In this paper, a new method for detection and classification of cardiac ailments from multilead electrocardiogram (MECG) is presented. The singular value decomposition (SVD) is used to convert the MECG data matrix into two unitary matrices (eigen matrices) and one diagonal matrix. According to clinical importance, first few atoms from the eigen matrices are selected. The root mean square error (RMSE) between the unitary matrices of both template MECG and analyzed MECG are used as diagnostic eigen error (DEE) features. The combination of singular values of analyzed MECG and DEE features are used as input to the least square support vector machine (LSSVM) classifier. The LSSVM detect the cardiac ailments such as myocardial infarction and hypertrophy. An average accuracy of 95.07% is found using LSSVM classifier with radial basis function (RBF) kernel and 5-fold cross-validation scheme.

Stressed speech analysis using sparse representation over temporal information based dictionary

Abstract: In this paper, a novel sparse representation over learned and exemplar dictionaries is explored to estimate the speech information of stressed speech. Stressed speech contains speech and stress informations. The acoustic variabilities are induced due to presence of stress information, which results in degradation of the performance of speech recognition system. In this work, the acoustic variabilities are reduced by representing both neutral and stressed speech in sparse domain with respect to the dictionaries, which contain speech information. K-SVD algorithm is used to learn the redundant dictionary using neutral speech. Exemplar dictionaries consist of mean vectors of GMM and mean vectors of Gaussian mixture density in each state of HMM, which are used to model the neutral speech. All the experiments in this work are done by parametrizing neutral and stressed speech as nonlinear (TEO-CB-Auto-Env) features. Experimental results indicate that speech information under stress conditions can be estimated efficiently when sparse representations of neutral and stressed speech are done over exemplar dictionaries, which is estimated using mean vectors of Gaussian mixture densities in each state of HMM i.e. time dependent features of neutral speech. A relative improvement in the percentage of word accuracy of 8.51% (62.14% to 67.43%) is achieved for speech under angry condition.

SVD Analysis on Reduced 3-Lead ECG Data

Abstract: This paper presents synthesis of Electrocardiogram (ECG) leads from reduced set of leads. The Singular Value Decomposition (SVD) is used to train subject-specific all desired leads for minimum of three beat periods. Then, in the testing phase, only 3-leads are used to reconstruct all other leads. The singular value matrix of the reduced 3-lead data is transformed to a higher dimension using a transform matrix. For evaluation purpose, the proposed method is applied to a publicly available database. It contains number of 12-lead ECG recordings with different cardiac patients data. After synthesis of ECG data, the performance of the method is measured using percent correlation present between the original and synthesized data.

Synthesis of 12-lead ECG from a reduced lead set using singular value decomposition

Abstract: This paper presents synthesis of 12-lead Electrocardiogram (ECG) from a reduced set of leads. A patient-specific 12-lead ECG is trained using the Singular Value Decomposition (SVD) for minimum of three beat periods. Then, in the testing stage, different reduced lead sets comprising of three leads are used to reconstruct all other leads. The singular value matrix of the reduced 3-lead data is transformed to a higher dimension using a transform matrix. For evaluation purpose, the proposed method is applied to a publicly available database. It contains number of 12-lead ECG recordings with different cardiac patients' data. The proposed method performs well for reconstruction of precordial leads in case of myocardial infarction. The percent correlation exceeds 97% for most of the precordial leads. After reconstruction of 12-lead ECG data, the performance of the method is measured using three measures namely percent correlation, PRD and WEDD between the original and the synthesized data.

Exploiting interchannel correlations in multichannel compressed sensing ECG systems

Abstract: Multichannel elctrocardiogram (MECG) signals are correlated both in spatial domain as well as in temporal domain and this correlation is stronger at multiscale levels (Fig 1) To exploit this correlation in compressed sensing (CS) based ECG tele-monitoring systems, a joint multiscale compressed sensing (JMCS) technique is proposed in this work CS is a novel signal acquisition/reconstruction paradigm that is proposed for addressing the power efficiency and complexity issues in wireless body area network (WBAN) enabled ECG tele-monitoring systems Here, JMCS is proposed to apply on jointly sparse subband signals from each channel instead of time domain MECG Since subband signals at each wavelet scale are more correlated, they share strong common support information and hence posses higher joint sparsity than MECG signals in time domain Joint acquisition and reconstruction of high frequency subband signals is formulated as a multi-measurement vector (MMV) problem To exploit the subband dependencies during the joint reconstruction a sparse Bayesian learning (SBL) based algorithm is employed which is known to be very efficient for finding joint sparse solution Significant performance gain in terms of diagnostic quality of reconstructed MECG is achieved at reduced number of measurements which directly translates into higher compression efficiency of the CS based WBAN systems

Quantifying Clinical Information in MECG Using Sample and Channel Convolution Matrices

Abstract: In this paper, a novel distortion measure is presented for quantifying loss of clinical information in multichannel electrocardiogram (MECG) signals. The proposed measure (SCPRD) is defined as the sum of percentage root mean square difference between magnitudes of convolution response of original and processed MECG signals. The convolution operation is performed with the help of proposed sample and channel convolution matrices. The SCPRD measure is compared with average wavelet energy diagnostic distortion (AWEDD) and multichannel PRD (MPRD) measures over different processing schemes such as multiscale principal component analysis (MSPCA) and multichannel empirical mode decomposition (MEMD)-based MECG compression and filtering. The normal and pathological MECG signals from the Physikalisch Technische Bundesanstalt (PTB) database is used in this work. The result shows that the proposed diagnostic distortion measure is effective to quantify the loss of clinical information in MECG signals.

Two-Dimensional Processing of Multichannel ECG Signals for Efficient Exploitation of Inter and Intra-Channel Correlation

Abstract: Electrocardiogram signals acquired through different channels from the body surface are termed as Multichannel ECG (MECG) signals They are obtained by projecting the same heart potential in different directions and hence share common information with each other In this work a new two-dimensional (2-D) approach is proposed for MECG signal processing in order to exploit the correlated structure between the channels efficiently Different channel data are arranged in a 2-D form giving them an image type arrangement and then 2-D discrete cosine transform (DCT) is applied in a blockwise manner over the whole data The 2-D processing of MECG data ensures the efficient utilization of both inter-lead correlation (across the columns) and intra-lead correlation (across the rows) Since neighboring ECG samples across the channels are more correlated due to slowly varying nature of ECGs, blockwise processing of MECG data gives an effective way to exploit this To quantify the performance of the proposed algorithm, it is evaluated on a compression platform Each block after DCT transformation is undergone through a uniform scale zero-zone quantizer and entropy encoder to get the compressed bit streams Performance metrics used are the compression ratio (CR) , and widely used distortion measure, root mean square difference (PRD)

Stressed Speech Recognition Using Similarity Measurement on Inner Product Space

Abstract: In this paper, similarity measurement on different inner product space approach is proposed for analysis of stressed speech The similarity is measured between neutral speech subspace and stressed speech subspace Cosine between neutral speech and stressed speech is taken as similarity measurement parameter It is asssumed that, speech and stress components of stressed speech are linearly related to each other Cosine between neutral and stressed speech multiples of stressed speech contains speech information of stressed speech Complement cosine (1-cosine) multiples of stressed speech is taken as stress component of stressed speech Neutral speech subspace is created by all neutral speech of the training database and stressed speech subspace contain stressed (angry, sad, lombard, happy) speech From experiment, it is observed that, stress information of stressed speech is not present in the complement cosine (1-cosine) times of stressed speech on different inner product space The linear relationship between speech and stress component of stressed speech exists only for some specific inner product space All the experiments are done using nonlinear (TEO-CB-Auto-Env) feature