Wearable and smartphone-based emotion recognition (WER) remains a challenging setting in affective computing, due to the notorious difficulty and bias associated with in-thewild label collection. The high inter-and intra-subject emotional variability motivates us to explore WER modeling through graph node classification in a limited resources learning scheme powered by Self-Supervised Learning (SSL) graph masking augmentation tasks. We employ a subgraph sampling approach during training, utilizing labeled and unlabeled data, along with supervised, semi-supervised, and SSL mechanisms in a multi-task inductive graph neural network architecture. Our evaluations on K-EmoPhone through leave-one-group-out cross-validation in the binary arousal and valence tasks yield average accuracy gains of 4.3% and 7.8%, compared to the full resource setting, utilizing only 20% and 25% of the labels, respectively. Our model analysis sheds light on the relation of SSL graph augmentations to emotional arousal and valence and justifies the approach of SSL-driven subgraph training for in-the-wild WER.

Obstructive Sleep Apnea-Hypopnea Syndrome (OSAHS) is a prevalent sleep disorder characterized by recurrent episodes of obstructed breathing due to the relaxation of muscles in the upper airway during sleep, often linked with neuromuscular and cardiovascular disorders. This study introduces a novel method using traditional machine learning classifiers and surface electromyography (SEMG) features extracted from motor units (MUs) decomposed from chin electromyography (EMG) signals to screen for OSA events in OSAHS subjects. SEMG features were extracted from individual MUs decomposed from chin EMG segments using a novel dataset. An apnea detection algorithm was designed to label these events for OSAHS subjects across sleep stages. Analysis of motor neuron firing patterns in OSAHS subjects revealed lower activation during OSA events and higher activation during non-OSA segments. Additionally, we evaluated the proposed system on a publicly available dataset, achieving a maximum accuracy of 72% for OSAHS subjects in the midlife phase age group (40–59 years) and 72.5% for subjects in the severe phase of OSAHS using Support Vector Machines (SVM). The random forest (RF) classifier demonstrated robust performance, achieving 97% accuracy, 93.2% sensitivity, 100% specificity, 100% precision, a 96.48% F1-score, and an area under the curve (AUC) of 0.996. This system facilitates early differentiation between OSA and non-OSA events, enabling timely intervention in the mild apnea phase to prevent progression to severe OSAHS. Moreover, it offers a convenient alternative to conventional polysomnography (PSG), enhancing diagnostic accessibility and clinical management.

ABSTRACT The growing availability of conversational data across multiple platforms has intensified interest in dynamic emotion recognition. Speech plays a pivotal role in shaping the emotional climate (EC) of peer conversations. We propose DeepBispec, the first framework to integrate deep bispectral image analysis with affect dynamics (AD) for speech‐based EC recognition. Bispectrum representations capture nonlinear and non‐Gaussian speech characteristics, while AD descriptors model temporal emotion fluctuations. Evaluated on K‐EmoCon, IEMOCAP and SEWA datasets, DeepBispec consistently improved EC classification performance. For example, on K‐EmoCon, arousal accuracy increased from 79.0% (bispectrum only) to 81.4% (with AD), while valence accuracy improved from 76.8% to 77.5%; similar trends were observed for IEMOCAP and SEWA. DeepBispec outperformed strong CNN, LSTM, and Transformer baselines, demonstrating robust cross‐lingual performance across seven languages. These findings highlight its potential for real‐world applications such as mental health monitoring, affect‐aware learning platforms and empathetic dialogue systems.

While machines struggle to cope with acoustical variability and noise, humans show remarkable robustness to recognize speech content under different conditions of environmental noise. The tonotopic organization of the spiral human cochlea has motivated the signal processing community for its superb frequency tuning capabilities. In this work, we design and evaluate a novel spiral cochlear cepstrum space, as a novel, directional feature engineering framework, using a cochlear transform approach, that results in tonotopically organized, orthogonal cochlear modes. Such cochlear modes are then transformed to the spiral cochlear cepstral space, yielding cochlear filterbank cepstral coefficients (CFCCs). As opposed to previous works that define the bio-inspired cepstral features based on Mel-, Equivalent Rectangular Bandwidth (ERB) or linear scales, we define the scaling based on the cochlear spiral geometry that spans from θ = 0° at the base to θ = 990° at the apex. We then compute the log function and the discrete cosine transform of the cochlear modes energy yielding spatially supported cepstral features along the spiral cochlear space, spaced by θ = 45°. We assess the impact of noise on the CFCCs and compare the performance to that of Mel-Frequency Cepstral Coefficients (MFCCs) and Gammatone Filterbank Cepstral Coefficients (GFCCs) using the NOIZEUS dataset. We report, for the first time, that the superiority of the CFCCs noise-robustness stems from the geometrical organization of the cochlea (i.e., its tonotopic map) when evaluated on speech signals contaminated with different noise conditions at different SNRs. The proposed CFCCs constitute a platform for a new class, bio-inspired and noise-robust feature extraction for many applications such as speaker recognition.

One in every four newborns suffers from congenital heart disease (CHD) that causes defects in the heart structure. The current gold-standard assessment technique, echocardiography, causes delays in the diagnosis owing to the need for experts who vary markedly in their ability to detect and interpret pathological patterns. Moreover, echo is still causing cost difficulties for low- and middle-income countries. Here, we developed a deep learning-based attention transformer model to automate the detection of heart murmurs caused by CHD at an early stage of life using cost-effective and widely available phonocardiography (PCG). PCG recordings were obtained from 942 young patients at four major auscultation locations, including the aortic valve (AV), mitral valve (MV), pulmonary valve (PV), and tricuspid valve (TV), and they were annotated by experts as absent, present, or unknown murmurs. A transformation to wavelet features was performed to reduce the dimensionality before the deep learning stage for inferring the medical condition. The performance was validated through 10-fold cross-validation and yielded an average accuracy and sensitivity of 90.23 % and 72.41 %, respectively. The accuracy of discriminating between murmurs' absence and presence reached 76.10 % when evaluated on unseen data. The model had accuracies of 70 %, 88 %, and 86 % in predicting murmur presence in infants, children, and adolescents, respectively. The interpretation of the model revealed proper discrimination between the learned attributes, and AV channel was found important (score 0.75) for the murmur absence predictions while MV and TV were more important for murmur presence predictions. The findings potentiate deep learning as a powerful front-line tool for inferring CHD status in PCG recordings leveraging early detection of heart anomalies in young people. It is suggested as a tool that can be used independently from high-cost machinery or expert assessment.

Major depressive disorder (MDD) is a chronic mental illness which affects people's well-being and is often detected at a later stage of depression with a likelihood of suicidal ideation. Early detection of MDD is thus necessary to reduce the impact, however, it requires monitoring vitals in daily living conditions. EEG is generally multi-channel and due to difficulty in signal acquisition, it is unsuitable for home-based monitoring, whereas, wearable sensors can collect single-channel ECG. Classical machine-learning based MDD detection studies commonly use various heart rate variability features. Feature generation, which requires domain knowledge, is often challenging, and requires computation power, often unsuitable for real time processing, MDDBranchNet is a proposed parallel-branch deep learning model for MDD binary classification from a single channel ECG which uses additional ECG-derived signals such as R-R signal and degree distribution time series of horizontal visibility graph. The use of derived branches was able to increase the model's accuracy by around 7%. An optimal 20-second overlapped segmentation of ECG recording was found to be beneficial with a 70% prediction threshold for maximum MDD detection with a minimum false positive rate. The proposed model evaluated MDD prediction from signal excerpts, irrespective of location (first, middle or last one-third of the recording), instead of considering the entire ECG signal with minimal performance variation stressing the idea that MDD phenomena are likely to manifest uniformly throughout the recording.

Hypertension is a serious medical condition that affects over a billion people worldwide. The proper management of disease progression requires an extended knowledge of the overall functional and structural changes in the whole body in response to the hypertension. Here, we propose HyperScore, an integrative and unified measure of hypertension progression relative to multi-organ and multi-modality clinical measurements and based on a semi-supervised machine learning (ML) approach. We developed the measure based on a large participating cohort from the UK Biobank database (n=27,099) with over 500 imaging and clinical variables from multiple modalities. The semi-supervised approach was developed based on the contrastive trajectory inference mechanism to provide a score that reflects the proximity of a participant to the disease state (range: 0–1). Modelling revealed that majority of hypertensive participants had scores above 0.25, whereas normotensives had scores below this threshold. The sensitivity and specificity were above 89%, with an area under the receiver operating characteristics of 96.4%. The modelling showed a stable performance when evaluating hidden testing sets on a 10-fold cross-validation scheme with nearly 0.1 error. There was a strong association (r2>0.6) between HyperScore and organs’ phenotypic patterns, especially for variables such as white matter hyperintensity and body mass index. This study is the first to potentiate ML-based modelling of hypertension progression from a multi-organ perspective, which could significantly aid in clinical decision making to save lives.

Signal multiscale decomposition (SMD) is an effective analysis for the identification of modal information in time-domain signals. So far, various SMD approaches, such as the Multiresolution Wavelet Transform (MWT), the Empirical Mode Decomposition (EMD), and the Variational Mode Decomosition (VMD) have been proposed. However, issues, such as mode mixing for signals with closelyspaced modes, have been identified. To confront such problems, we propose here a novel spatial auditory decomposition framework for non-stationary signals, namely the Cochlear Decomposition (CD). CD is inspired by the biological rules of the spiral human cochlea and it is built upon the concept of ‘place-pitch’ or the tonotopic organization of the spiral cochlea. The new insight here is to formulate a set of basis functions, namely cochlear wavelets, whose dilation factors are determined by their angular position on the cochlear spiral (i.e., coupled rotation and dilation). Under proper parameterization, iterative application of spatial filters eventually results in signal’s mono-components, each arising from specific angular positions along the spiral cochlea, with high time and frequency localization. The performance of the proposed CD is validated via synthetic acoustic signals and real non-stationary speech signals analysis. The analysis results show that CD outperforms the performance of MWT, EMD, and VMD, exhibiting, at the same time, high noise robustness. We also show that the CD disentangles lowfrequency temporal modulations of sounds, supporting perceptual phenomena in both speech and music. Clearly, CD paves the way for higher-level speech perception deep learning models and also for efficient cochlear implants design. Future work to incorporate the nonlinearity of the cochlea into a binaural hearing framework is underway.

Fetal cardiac monitoring is very helpful in the early detection of the potential risk of fetal cardiac abnormalities, which enables prompt preventative care and ensures safe births. As a result, it is crucial to regularly check on the embryonic heart. Methods of non-invasively fetal ECG extraction from maternal abdominal ECG signal are thoroughly discussed. Although fetal signals are generally obscured by maternal ECG signals and noise, extracting a clean fetal ECG is a significant difficulty. The majority of techniques for fetal ECG extraction include many extraction steps. We describe a unique method for splitting a single-channel maternal abdominal ECG into maternal and fetus ECG employing two parallel U-nets with transformer encoding, which we refer to as W-NEt TRansformers (W-NETR). Due to its enhanced capacity to simulate remote interactions and capture global context, the suggested pipeline utilizes the self-attention mechanism of the transformer. We tested the proposed pipeline on synthetic and real datasets and outperformed the current state-of-the-art deep learning models. The proposed model achieved the best results on both datasets for QRS detection precision, recall, and F1 scores. More specifically, it achieved F1 score of 99.88% and 98.9% on the real ADFECGDB and PCDB datasets, respectively. These encouraging results highlight the suggested W-NETR's effectiveness in precisely extracting the fetal ECG, which was achieved with high SSIM and PSNR values in the results. This provides the bed set for long-term maternal and fetal monitoring via portable devices as the proposed system performs real-time execution.

The automated recognition of human emotions plays an important role in developing machines with emotional intelligence. Major research efforts are dedicated to the development of emotion recognition methods. However, most of the affective computing models are based on images, audio, videos and brain signals. Literature lacks works that focus on utilizing only peripheral signals for emotion recognition (ER), which can be ideally implemented in daily life settings. Therefore, this paper present a framework for ER on the arousal and valence space, based on using multi-modal peripheral signals. The data used in this work were collected during a debate between two people using wearable devices. The emotions of the participants were rated by multiple raters and converted into classes in correspondence to the arousal and valence space. The use of a dynamic threshold for ratings conversion was investigated. An ER model is proposed that uses a Long Short-Term Memory (LSTM)-based architecture for classification. The model uses heart rate (HR), temperature (T), and electrodermal activity (EDA) signals as its inputs with emotional cues. Additionally, a post-processing prediction mechanism is introduced to enhance the recognition performance. The model is implemented to study the use of individual and different combinations of the peripheral signals, as well as utilizing annotations from different ratings. Additionally, it is employed for classification of valence and arousal in an independent and combined fashion, under subject dependent and independent scenarios. The experimental results have justified the efficient performance of the proposed framework, achieving classification accuracy 96% and 93% for the independent and combined classification scenarios, accordingly. The comparison of the achieved performance against the baseline methods shows the superiority of the proposed framework and the ability to recognize arousal-valance levels with high accuracy from peripheral signals, in real-life scenarios.

Long-term, continuous physiological recordings are currently being intensely investigated for tracking emotions. Emotional valence has been of more interest due to its relevance to cardiac and neurophysiological disease. In this research, multiple configurable convolutional neural networks (CNNs) were developed for different image-encoding techniques used as their input. Ensemble classification was then used to achieve a combined performance of the multiple CNNs by training a simple support vector machine (SVM) classifier using the last output layers of the CNNs as its input. Valence-labelled signals from the heart rate (HR) recorded using a wearable sensor from a wristband in a daily setting for one week from 80 participants were used for the image transforms. Accuracies of more than 91% were achieved with the classification ensembling, showing an improvement of the binary classification of emotional valence by more than 19% compared to using CNNs on their own.

OBJECTIVES: The purpose of this study was to set an optimal fit of the estimated LVEF at hourly intervals from 24-hour ECG recordings and compare it with the fit based on two gold-standard guidelines. METHODS: Support vector regression (SVR) models were applied to estimate LVEF from ECG derived heart rate variability (HRV) data in one-hour intervals from 24-hour ECG recordings of patients with either preserved, mid-range, or reduced LVEF, obtained from the Intercity Digital ECG Alliance (IDEAL) study. A step-wise feature selection approach was used to ensure the best possible estimations of LVEF levels. RESULTS: The experimental results have shown that the lowest Root Mean Square Error (RMSE) between the original and estimated LVEF levels was during 3-4 am, 5-6 am and 6-7 pm. CONCLUSION: The observations suggest these hours as possible times for intervention and optimal treatment outcomes. In addition, LVEF classifications following the ACCF/AHA guidelines leads to a more accurate assessment of mid-range LVEF. SIGNIFICANCE: This study paves the way to explore the use of HRV features in the prediction of LVEF percentages as an indicator of disease progression, which may lead to an automated classification process for CAD patients.

Electrocardiography (ECG) represents the hearts electrical activity and has features such as QRS complex, P-wave and T-wave that provide critical clinical information for detection and prediction of cardiac diseases. This paper presents a novel ECG processing architecture for the prediction of ventricular arrhythmia (VA). The architecture implements a novel ECG feature extraction which is optimized for ultra-low power applications. The architecture is based on Curve Length Transform (CLT) for the detection of QRS complex and Discrete Wavelet Transform (DWT) for the delineation of TP waves. Features extracted from two consecutive ECG cycles are used to set innovative parameters for VA prediction up to 3 hours before VA onset. Two databases of the heart signal recordings from the American Heart Association (AHA) and the MIT PhysioNet were used as training, test and validation sets to evaluate the performance of the proposed system.

In this study, a linear parametric modeling technique was applied to model ventricular repolarization (VR) dynamics. Three features were selected from the surface ECG recordings to investigate the changes in VR dynamics in healthy and cardiac autonomic neuropathy (CAN) participants with diabetes including heart rate variability (calculated from RR intervals), repolarization variability (calculated from QT intervals), and respiration [calculated by ECG-derived respiration (EDR)]. Surface ECGs were recorded in a supine resting position from 80 age-matched participants (40 with no cardiac autonomic neuropathy (NCAN) and 40 with CAN). In the CAN group, 25 participants had early/subclinical CAN (ECAN) and 15 participants were identified with definite/clinical CAN (DCAN). Detecting subclinical CAN is crucial for designing an effective treatment plan to prevent further cardiovascular complications. For CAN diagnosis, VR dynamics was analyzed using linear parametric autoregressive bivariate (ARXAR) and trivariate (ARXXAR) models, which were estimated using 250 beats of derived QT, RR, and EDR time series extracted from the first 5 min of the recorded ECG signal. Results showed that the EDR-based models gave a significantly higher fitting value (p < 0.0001) than models without EDR, which indicates that QT-RR dynamics is better explained by respiratory-information-based models. Moreover, the QT-RR-EDR model fitting values gradually decreased from the NCAN group to ECAN and DCAN groups, which indicate a decoupling of QT from RR and the respiration signal with the increase in severity of CAN. In this study, only the EDR-based model significantly distinguished ECAN and DCAN groups from the NCAN group (p < 0.05) with large effect sizes (Cohen's d > 0.75) showing the effectiveness of this modeling technique in detecting subclinical CAN. In conclusion, the EDR-based trivariate QT-RR-EDR model was found to be better in detecting the presence and severity of CAN than the bivariate QT-RR model. This finding also establishes the importance of adding respiratory information for analyzing the gradual deterioration of normal VR dynamics in pathological conditions, such as diabetic CAN.

Electromechanical coupling of the fetal heart can be evaluated noninvasively using doppler ultrasound (DUS) signal and fetal electrocardiography (fECG). In this study, an efficient model is proposed using K-means clustering and hybrid Support Vector Machine-Hidden Markov Model (SVM-HMM) modeling techniques. Opening and closing of the cardiac valves were detected from peaks in the high frequency component of the DUS signal decomposed by wavelet analysis. It was previously proposed to automatically identify the valve motion by hybrid SVM-HMM based on the amplitude and timing of the peaks. However, in the present study, six patterns were identified for the DUS components which were actually variable on a beat-to-beat basis and found to be different for the early gestation (16-32 weeks), compared to the late gestation fetuses (36-41 weeks). The amplitude of the peaks linked to the valve motion was different across the six patterns and this affected the precision of valve motion identification by the previous hybrid SVM-HMM method. Therefore in the present study, clustering of the DUS components based on K-means was proposed and the hybrid SVM-HMM was trained for each cluster separately. The valve motion events were consequently identified more efficiently by beat-to-beat attribution of the DUS component peaks. Applying this method, more than 98.6% of valve motion events were beat-to-beat identified with average precision and recall of 83.4% and 84.2% respectively. It was an improvement compared to the hybrid method without clustering with average precision and recall of 79.0% and 79.8%. Therefore, this model would be useful for reliable screening of fetal wellbeing.

Angiotensin converting enzyme (ACE) polymorphism has been shown to be important in hypertension progression and also in diabetes complications, especially associated with heart disease. Heart rate variability (HRV) is an established measure for classification of autonomic function regulating heart rate, based on the interbeat interval time series derived from a raw ECG recording. Results of this paper show that the length (number of interbeat intervals) and preprocessing of the tachogram affect the HRV analysis outcome. The comparison was based on tachogram lengths of 250, 300, 350, and 400 RR-intervals and five preprocessing approaches. An automated adaptive preprocessing method for the heart rate biosignal and tachogram length of 400 interbeat intervals provided the best classification. HRV results differed for the Type 2 Diabetes Mellitus (T2DM) group between the I/I genotype and the I/D and D/D genotypes, whereas for controls there was no significant difference in HRV between genotypes. Selecting an appropriate length of recording and automated preprocessing has confirmed that there is an effect of ACE polymorphism including the I/I genotype and that I/I should not be combined with I/D genotype in determining the extent of autonomic modulation of the heart rate.

This paper presents the design of a fully integrated electrocardiogram (ECG) signal processor (ESP) for the prediction of ventricular arrhythmia using a unique set of ECG features and a naive Bayes classifier. Real-time and adaptive techniques for the detection and the delineation of the P-QRS-T waves were investigated to extract the fiducial points. Those techniques are robust to any variations in the ECG signal with high sensitivity and precision. Two databases of the heart signal recordings from the MIT PhysioNet and the American Heart Association were used as a validation set to evaluate the performance of the processor. Based on application-specified integrated circuit (ASIC) simulation results, the overall classification accuracy was found to be 86% on the out-of-sample validation data with 3-s window size. The architecture of the proposed ESP was implemented using 65-nm CMOS process. It occupied 0.112- ${\rm mm}^{2}$ area and consumed 2.78- $\mu \text{W}$ power at an operating frequency of 10 kHz and from an operating voltage of 1 V. It is worth mentioning that the proposed ESP is the first ASIC implementation of an ECG-based processor that is used for the prediction of ventricular arrhythmia up to 3 h before the onset.

ECG intervals such as QRS, QT and PR provide significant information and are widely used as clinical parameters for diagnosing cardiac diseases. This paper presents a novel QRS detection technique based on Curve Length Transform (CLT) and a refined delineation of P-wave and T-wave using Discrete Wavelet Transform (DWT). The proposed technique was verified using the PhysioNet database. The QRS detection achieved a sensitivity of 98.59% and a positive predictivity of 97.86%. The QRS duration, QT interval and PR interval had a mean error of -1.56± 28.8ms, -5.39± 42.4ms and 0.86± 40.3ms respectively. The proposed algorithm is computationally efficient and is simpler to implement in hardware, hence, will lead to a faster execution time, smaller design area and consequently low power consumption.

Respiratory events during sleep induce cortical arousals and manifest changes in autonomic markers in sleep disorder breathing (SDB). Finger photoplethysmography (PPG) has been shown to be a reliable method of determining sympathetic activation. We hypothesize that changes in PPG signals are sufficient to predict the occurrence of respiratory-event-related cortical arousal. In this study, we develop a respiratory arousal detection model in SDB subjects by using PPG features. PPG signals from 10 SDB subjects (9 male, 1 female) with age range 43-75 years were used in this study. Time domain features of PPG signals, such as 1) PWA--pulse wave amplitude, 2) PPI--peak-to-peak interval, and 3) Area--area under peak, were used to detect arousal events. In this study, PWA and Area have shown better performance (higher accuracy and lower false rate) compared to PPI features. After investigating possible groupings of these features, combination of PWA and Area (PWA + Area) was shown to provide better accuracy with a lower false detection rate in arousal detection. PPG-based arousal indexes agreed well across a wide range of decision thresholds, resulting in a receiver operating characteristic with an area under the curve of 0.91. For the decision threshold (PC(thresh) = 25%) chosen for the final analyses, a sensitivity of 68.1% and a specificity of 95.2% were obtained. The results showed an accuracy of 84.68%, 85.15%, 86.93%, and 50.79% with a false rate of 21.80%, 55.41%, 64.78%, and 50.79% at PC(thresh) = 25% or PPI, PWA, Area , and PWA + Area features, respectively. This indicates that combining PWA and Area features reduced the false positive rate without much affecting the sensitivity of the arousal detection system. In conclusion, the PPG-based respiratory arousal detection model is a simple and promising alternative to the conventional electroencephalogram (EEG)-based respiratory arousal detection system.

In this paper, a new noninvasive method is proposed for automated estimation of fetal cardiac intervals from Doppler Ultrasound (DUS) signal. This method is based on a novel combination of empirical mode decomposition (EMD) and hybrid support vector machines-hidden Markov models (SVM/HMM). EMD was used for feature extraction by decomposing the DUS signal into different components (IMFs), one of which is linked to the cardiac valve motions, i.e. opening (o) and closing (c) of the Aortic (A) and Mitral (M) valves. The noninvasive fetal electrocardiogram (fECG) was used as a reference for the segmentation of the IMF into cardiac cycles. The hybrid SVM/HMM was then applied to identify the cardiac events, based on the amplitude and timing of the IMF peaks as well as the sequence of the events. The estimated timings were verified using pulsed doppler images. Results show that this automated method can continuously evaluate beat-to-beat valve motion timings and identify more than 91% of total events which is higher than previous methods. Moreover, the changes of the cardiac intervals were analyzed for three fetal age groups: 16-29, 30-35, and 36-41 weeks. The time intervals from Q-wave of fECG to Ac (Systolic Time Interval, STI), Ac to Mo (Isovolumic Relaxation Time, IRT), Q-wave to Ao (Preejection Period, PEP) and Ao to Ac (Ventricular Ejection Time, VET) were found to change significantly ( ) across these age groups. In particular, STI, IRT, and PEP of the fetuses with 36-41 week were significantly ( ) different from other age groups. These findings can be used as sensitive markers for evaluating the fetal cardiac performance.

Cardiovascular autonomic neuropathy (CAN) has been frequently postulated to increase susceptibility to ventricular arrhythmias and sudden cardiac death in diabetic patients. The relation between the progression of CAN in diabetes and ventricular repolarization remains to be fully described. Therefore, this study examined QT interval variability and heart rate interbeat variability to identify any alterations of cardiac repolarization in diabetic patients in relation to severity of CAN. Seventy control participants without (CAN-) and 74 patients with CAN (CAN+) were enrolled in this study. Among 74 CAN + patients, 62 are early CAN + (eCAN +) , and 12 are definite CAN + (dCAN +) according to autonomic nervous system function tests as described by Ewing. The results showed that the QT variability index (QTVI) was significantly higher and positive in the dCAN + (0.51 ±1.32) group than in the eCAN + (-0.39 ±0.91) and CAN - (-0.54 ±0.72) groups. The QT variability to heart-rate variability ratio provides a measure of the balance between QT and heart interbeat variability. QTVI was more sensitive in identifying disease progression at all stages. Our study supports the hypothesis that QTVI could be used as a clinical test to identify early CAN and as a marker of CAN progression in diabetic patients and may help physicians in determining the best therapeutic strategy for these patients.

Obstructive sleep apnea or hypopnea causes a pause or reduction in airflow with continuous breathing effort. The aim of this study is to identify individual apnea and hypopnea events from normal breathing events using wavelet-based features of 5-s ECG signals (sampling rate = 250 Hz) and estimate the surrogate apnea index (AI)/hypopnea index (HI) (AHI). Total 82,535 ECG epochs (each of 5-s duration) from normal breathing during sleep, 1638 ECG epochs from 689 hypopnea events, and 3151 ECG epochs from 1862 apnea events were collected from 17 patients in the training set. Two-staged feedforward neural network model was trained using features from ECG signals with leave-one-patient-out cross-validation technique. At the first stage of classification, events (apnea and hypopnea) were classified from normal breathing events, and at the second stage, hypopneas were identified from apnea. Independent test was performed on 16 subjects' ECGs containing 483 hypopnea and 1352 apnea events. The cross-validation and independent test accuracies of apnea and hypopnea detection were found to be 94.84% and 76.82%, respectively, for training set, and 94.72% and 79.77%, respectively, for test set. The Bland-Altman plots showed unbiased estimations with standard deviations of +/- 2.19, +/- 2.16, and +/- 3.64 events/h for AI, HI, and AHI, respectively. Results indicate the possibility of recognizing apnea/hypopnea events based on shorter segments of ECG signals.

Obstructive sleep apnea syndrome (OSAS) is associated with cardiovascular morbidity as well as excessive daytime sleepiness and poor quality of life. In this study, we apply a machine learning technique [support vector machines (SVMs)] for automated recognition of OSAS types from their nocturnal ECG recordings. A total of 125 sets of nocturnal ECG recordings acquired from normal subjects (OSAS - ) and subjects with OSAS (OSAS +), each of approximately 8 h in duration, were analyzed. Features extracted from successive wavelet coefficient levels after wavelet decomposition of signals due to heart rate variability (HRV) from RR intervals and ECG-derived respiration (EDR) from R waves of QRS amplitudes were used as inputs to the SVMs to recognize OSAS +/- subjects. Using leave-one-out technique, the maximum accuracy of classification for 83 training sets was found to be 100% for SVMs using a subset of selected combination of HRV and EDR features. Independent test results on 42 subjects showed that it correctly recognized 24 out of 26 OSAS + subjects and 15 out of 16 OSAS - subjects (accuracy = 92.85%; Cohen's kappa value of 0.85). For estimating the relative severity of OSAS, the posterior probabilities of SVM outputs were calculated and compared with respective apnea/hypopnea index. These results suggest superior performance of SVMs in OSAS recognition supported by wavelet-based features of ECG. The results demonstrate considerable potential in applying SVMs in an ECG-based screening device that can aid a sleep specialist in the initial assessment of patients with suspected OSAS.

The consequence of tripping and falling in the elderly population is serious because of the life threatening fractures which occur and the high medical costs incurred. Recently, the minimum toe clearance (MTC) has been employed in gait analysis as a sensitive gait variable for early detection of elderly people at risk of falling. In previous work, we successfully applied statistical and wavelet analysis methods with Support Vector Machines (SVM) to model the risk of tripping in the elderly. In this work, we propose to model the MTC time series as a wide based stationary random signal using the autoregressive (AR) process. Initially, it was found that a fourth order AR model constructed from 512 MTC samples per subject on 23 subjects completely modelled the balance impaired gait (pathological) from normal gait. However, when the number of MTC samples were reduced to 32, the two groups became inseparable. We then proposed a hybrid system consisting of a SVM classifier with AR model coefficients as input features to separate the two classes. It was found that SVMs with linear and Gaussian kernels produced 100% leave one out accuracies without the need for prior feature selection algorithms. In contrast, SVM models built previously from the best set of wavelet features produced only 86.95% leave one out accuracies. These results suggest that pathological gait is best modelled by the AR process if sufficient MTC data is available. In the case of shorter MTC data, the AR model still provides powerful and robust discriminative features which can be used by the SVM to detect elderly people at risk of falling.