Udantha R. Abeyratne

Higher order spectra based deconvolution of ultrasound images

We address the problem of improving the spatial resolution of ulrasound images through blind deconvolution. The ultrasound image formation process in the RF domain can be expressed as a spatio-temporal convolution between the tissue response and the ultrasonic system response, plus additive noise. Convolutional components of the dispersive attenuation and aberrations introduced by propagating through the object being imaged are also incorporated in the ultrasonic system response. Our goal is to identify and remove the convolutional distortion in order to reconstruct the tissue response, thus enhancing the diagnostic quality of the ultrasonic image. Under the assumption of an independent, identically distributed, zero-mean, non-Gaussian tissue response, we were able to estimate distortion kernels using bicepstrum operations on RF data. Separate 1D distortion kernels were estimated corresponding to axial and lateral image lines and used in the deconvolution process. The estimated axial kernels showed similarities to the experimentally measured pulse-echo wavelet of the imaging system. Deconvolution results from B-scan images obtained with clinical imaging equipment showed a 2.5-5.2 times gain in lateral resolution, where the definition of the resolution has been based on the width of the autocovariance function of the image. The gain in axial resolution was found to be between 1.5 and 1.9.

Silence-breathing-snore classification from snore-related sounds.

Obstructive sleep apnea (OSA) is a highly prevalent disease in which upper airways are collapsed during sleep, leading to serious consequences. Snoring is the earliest symptom of OSA, but its potential in clinical diagnosis is not fully recognized yet. The first task in the automatic analysis of snore-related sounds (SRS) is to segment the SRS data as accurately as possible into three main classes: snoring (voiced non-silence), breathing (unvoiced non-silence) and silence. SRS data are generally contaminated with background noise. In this paper, we present classification performance of a new segmentation algorithm based on pattern recognition. We considered four features derived from SRS to classify samples of SRS into three classes. The features--number of zero crossings, energy of the signal, normalized autocorrelation coefficient at 1 ms delay and the first predictor coefficient of linear predictive coding (LPC) analysis--in combination were able to achieve a classification accuracy of 90.74% in classifying a set of test data. We also investigated the performance of the algorithm when three commonly used noise reduction (NR) techniques in speech processing--amplitude spectral subtraction (ASS), power spectral subtraction (PSS) and short time spectral amplitude (STSA) estimation--are used for noise reduction. We found that noise reduction together with a proper choice of features could improve the classification accuracy to 96.78%, making the automated analysis a possibility.

Multi-feature snore sound analysis in obstructive sleep apnea-hypopnea syndrome.

Snoring is the most common symptom of obstructive sleep apnea hypopnea syndrome (OSAHS), which is a serious disease with high community prevalence. The standard method of OSAHS diagnosis, known as polysomnography (PSG), is expensive and time consuming. There is evidence suggesting that snore-related sounds (SRS) carry sufficient information to diagnose OSAHS. In this paper we present a technique for diagnosing OSAHS based solely on snore sound analysis. The method comprises a logistic regression model fed with snore parameters derived from its features such as the pitch and total airway response (TAR) estimated using a higher order statistics (HOS)-based algorithm. Pitch represents a time domain characteristic of the airway vibrations and the TAR represents the acoustical changes brought about by the collapsing upper airways. The performance of the proposed method was evaluated using the technique of K-fold cross validation, on a clinical database consisting of overnight snoring sounds of 41 subjects. The method achieved 89.3% sensitivity with 92.3% specificity (the area under the ROC curve was 0.96). These results establish the feasibility of developing a snore-based OSAHS community-screening device, which does not require any contact measurements.

Mixed-phase modeling in snore sound analysis

Obstructive sleep apnea (OSA) is a highly prevalent disease in which upper airways are collapsed during sleep, leading to serious consequences. The gold standard of diagnosis, called Polysomnography (PSG), requires a full-night hospital stay connected to over 15 channels of measurements requiring physical contact with sensors. PSG is expensive and unsuited for community screening. Snoring is the earliest symptom of OSA, but its potential in OSA diagnosis is not fully recognized yet. In this paper, we propose a novel model for SRS as the response of a mixed-phase system (total airways response, TAR) to a source excitation at the input. The TAR/source model is similar to the vocal tract/source model in speech synthesis, and is capable of capturing acoustical changes brought about by the collapsing upper airways in OSA. We propose an algorithm based on higher-order-spectra (HOS) to jointly estimate the source and TAR, preserving the true phase characteristics of the latter. Working on a clinical database of signals, we show that TAR is indeed a mixed-phased signal and second-order statistics cannot fully characterize it. Night-time speech sounds can corrupt snore recordings and pose a challenge to snore based OSA diagnosis. We show that the TAR could be used to detect speech segments embedded in snores, and derive features to diagnose OSA via non-contact, low-cost instrumentation holding potential for a community screening device.

Artificial neural networks for source localization in the human brain

SummarySource localization in the brain remains an ill-posed problem unless further constraints about the type of sources and the head model are imposed. Human head is modeled in various ways depending critically on the computing power available and/or the required level of accuracy. Sophisticated and truly representative models may yield more accurate results in general, but at the cost of prohibitively long computer times and huge memory requirements. In conventional source localization techniques, solution source parameters are taken as those which minimize an index of performance, defined relative to the model-generated and clinically measured voltages. We propose the use of a neural network in the place of commonly employed minimization algorithms such as the Simplex Method and the Marquardt algorithm, which are iterative and time consuming. With the aid of the error-backpropagation technique, a neural network is trained to compute source parameters, starting from a voltage set measured on the scalp. Here we describe the methods of training the neural network and investigate its localization accuracy. Based on the results of extensive studies, we conclude that neural networks are highly feasible as source localizers. A trained neural networks independence of localization speed from the head model, and the rapid localization ability, makes it possible to employ the most complex head model with the ease of the simplest model. No initial parameters need to be guessed in order to start the calculation, implying a possible automation of the entire localization process. One may train the network on experimental data, if available, thereby possibly doing away with head models.

Pitch-jitter analysis of snoring sounds for the diagnosis of sleep apnea

Obstructive Sleep Apnea (OSA) is a disease in which airways involuntarily collapse during sleep, leading to serious consequences. About 10% of snorers suffer from OSA, unknown to them, nevertheless requiring medical attention. The current standard of diagnosis for OSA, polysomnography (PSG), requires that the patients spend one full day in a hospital, wired to a multitude of instruments. PSG is complicated, expensive, and unsuitable for mass screening of the population. OSA is commonly accompanied by snoring. Even though snoring carries vital information on the state of the airways, it has rarely been used in diagnosing OSA. In this paper, we present a mathematical model for snoring, and illustrate its usefulness in diagnosing OSA. We exploit similarities and differences between speech and snoring signals to separate the two, and, provide new features to diagnose OSA at low cost. Via experiments carried out in a hospital sleep-laboratory, we illustrate the importance of using noise reduction techniques to acquire snoring data with sufficient integrity.

Higher order versus second order statistics in ultrasound image deconvolution

We recently proposed a method for the estimation of imaging distortions associated with ultrasound images, based on the higher-order statistics (HOS) of radio frequency data. In this correspondence, we utilize the HOS-based estimated distortions to deconvolve ultrasound images of the breast. We also estimate imaging distortions based on the second-order statistics (SOS) of radio frequency ultrasound data and subsequently utilize them to deconvolve the same breast images. Both subjective and objective measures suggest that deconvolution with HOS-based distortion estimates led to significantly higher resolution gains as compared to the gains achieved when SOS-based distortion estimates were used.

Normal probability testing of snore signals for diagnosis of obstructive sleep apnea

Obstructive Sleep Apnea (OSA) is a highly prevalent disease in which upper airways are collapsed during sleep, leading to serious consequences. The standard method of OSA diagnosis is known as Polysomnography (PSG), which requires an overnight stay in a specifically equipped facility, connected to over 15 channels of measurements. PSG requires (i) contact instrumentation and, (ii) the expert human scoring of a vast amount of data based on subjective criteria. PSG is expensive, time consuming and is difficult to use in community screening or pediatric assessment. Snoring is the most common symptom of OSA. Despite the vast potential, however, it is not currently used in the clinical diagnosis of OSA. In this paper, we propose a novel method of snore signal analysis for the diagnosis of OSA. The method is based on a novel feature that quantifies the non-Gaussianity of individual episodes of snoring. The proposed method was evaluated using overnight clinical snore sound recordings of 86 subjects. The recordings were made concurrently with routine PSG, which was used to establish the ground truth via standard clinical diagnostic procedures. The results indicated that the developed method has a detectability accuracy of 97.34%.

Snore Signal Enhancement and Activity Detection via Translation-Invariant Wavelet Transform

Acoustical properties of snores have been widely studied as a potentially cost-effective and reliable alternative to diagnosing obstructive sleep apnea (OSA), with a common recognition that the diagnostic accuracy depends heavily on the snore signal quality and intelligibility. This paper proposes a novel preprocessing system that performs two critical tasks concurrently in a translation-invariant wavelet transform domain. These tasks include enhancement of snore signals via a level-correlation-dependent (LCD) threshold, and identification of snore presence through a snore activity (SA) detector. Various experiments were conducted to warrant the robustness of the system in terms of theoretical statistics quality, signal-to-noise ratio, mean opinion score, and clinical usefulness in detecting OSA. Results indicate that the proposed LCD threshold and SA detector are highly comparable to the existing denoising methodologies using level-dependent threshold and segmentation approaches using short-time energy and zero-crossing rate, yielding the best results in all the experiments. Given the strong initial performance of the proposed preprocessing system for snore signals, continued exploration in this direction could potentially lead to additional improvement in signal integrity, thereby increasing the diagnostic accuracy for OSA.

Obstructive sleep apnea screening by integrating snore feature classes

Obstructive sleep apnea (OSA) is a serious sleep disorder with high community prevalence. More than 80% of OSA suffers remain undiagnosed. Polysomnography (PSG) is the current reference standard used for OSA diagnosis. It is expensive, inconvenient and demands the extensive involvement of a sleep technologist. At present, a low cost, unattended, convenient OSA screening technique is an urgent requirement. Snoring is always almost associated with OSA and is one of the earliest nocturnal symptoms. With the onset of sleep, the upper airway undergoes both functional and structural changes, leading to spatially and temporally distributed sites conducive to snore sound (SS) generation. The goal of this paper is to investigate the possibility of developing a snore based multi-feature class OSA screening tool by integrating snore features that capture functional, structural, and spatio-temporal dependences of SS. In this paper, we focused our attention to the features in voiced parts of a snore, where quasi-repetitive packets of energy are visible. Individual snore feature classes were then optimized using logistic regression for optimum OSA diagnostic performance. Consequently, all feature classes were integrated and optimized to obtain optimum OSA classification sensitivity and specificity. We also augmented snore features with neck circumference, which is a one-time measurement readily available at no extra cost. The performance of the proposed method was evaluated using snore recordings from 86 subjects (51 males and 35 females). Data from each subject consisted of 6-8 h long sound recordings, made concurrently with routine PSG in a clinical sleep laboratory. Clinical diagnosis supported by standard PSG was used as the reference diagnosis to compare our results against. Our proposed techniques resulted in a sensitivity of 93±9% with specificity 93±9% for females and sensitivity of 92±6% with specificity 93±7% for males at an AHI decision threshold of 15 events/h. These results indicate that our method holds the potential as a tool for population screening of OSA in an unattended environment.