While the negative effects of alcohol on driving performance are undisputed, it is unclear how driver attention, eye movements and visual information sampling are affected by alcohol consumption. A simulator study with 35 participants was conducted to investigate whether and how a drivers level of attention is related to self-paced non-driving related task (NDRT)-engagement and tactical aspects of undesirable driver behaviour under increasing levels of breath alcohol concentration (BrAC) up to 1.0 %o. Increasing BrAC levels lead to more frequent speeding, short time headways and weaving, and higher NDRT engagement. Instantaneous distraction events become more frequent, with more and longer glances to the NDRT, and a general decline in visual attention to the forward roadway. With alcohol, the compensatory behaviour that is typically seen when drivers engage in NDRTs did not appear. These findings support the theory that alcohol reduces the ability to shift attention between multiple tasks. To conclude, the independent reduction in safety margins in combination with impaired attention and an increased willingness to engage in NDRTs is likely the reason behind increased crash risk when driving under the influence of alcohol.

Driver fatigue is a contributing factor in about 20% of all fatal road crashes worldwide. Countermeasures are urgently needed and one of the most promising and currently available approaches for that are in-vehicle systems for driver fatigue detection. The main objective of this paper is to present a video-based driver sleepiness detection system set up as a two-stage model with (1) a generic deep feature extraction module combined with (2) a personalised sleepiness detection module. The approach was designed and evaluated using data from 13 drivers, collected during naturalistic driving conditions on a motorway in Sweden. Each driver performed one 90-minute driving session during daytime (low sleepiness condition) and one session during night-time (high sleepiness condition). The sleepiness detection model outputs a continuous output representing the Karolinska Sleepiness Scale (KSS) scale from 1-9 or a binary decision as alert (defined as KSS 1-6) or sleepy (defined as KSS 7-9). Continuous output modelling resulted in a mean absolute error (MAE) of 0.54 KSS units. Binary classification of alert or sleepy showed an accuracy of 92% (sensitivity = 91.7%, specificity = 92.3%, F1 score = 90.4%). Without personalisation, the corresponding accuracy was 72%, while a standard fatigue detection PERCLOS-based baseline method reached an accuracy of 68% on the same dataset. The developed real-time sleepiness detection model can be used in the management of sleepiness/fatigue by detecting precursors of severe fatigue, and ultimately reduce sleepiness-related road crashes by alerting drivers before high levels of fatigue are reached.

Hours of service regulations govern the working hours of commercial motor vehicle drivers, but these regulations may become more flexible as highly automated vehicles have the potential to afford periods of in-cab rest or even sleep while the vehicle is moving. A prerequisite is robust continuous monitoring of when the driver is resting (to account for reduced time on task) or sleeping (to account for the reduced physiological drive to sleep). The overall aims of this paper are to raise a discussion of whether it is possible to obtain successful rest during automated driving, and to present initial work on a hypothetical data driven algorithm aimed to estimate if it is possible to gain driving time after resting under fully automated driving. The presented algorithm consists of four central components, a heart rate-based relaxation detection algorithm, a camera-based sleep detection algorithm, a fatigue modelling component taking time awake, time of day and time on task into account, and a component that estimates gained driving time. Real-time assessment of driver fitness is complicated, especially when it comes to the recuperative value of in-cab sleep and rest, as it depends on sleep quality, time of day, homeostatic sleep pressure and on the activities that are carried out while resting. The monotony that characterizes for long-haul truck driving is clearly interrupted for a while, but the long-term consequences of extended driving times, including user acceptance of the key stakeholders, requires further research.

This editorial serves as an extended introduction to the Special Issue on Technologies for Risk Mitigation and Support of Impaired Drivers. It gives the context to recent advances in assisted and automated driving and the new challenges that arise when modern technology meets human users. The Special Issue focuses on the development of robust sensors and detection algorithms for driver state monitoring of fatigue, stress, and inattention, and on the development of personalized multimodal, user-oriented, and adaptive information, warning, actuation, and handover strategies. A summary of more recent developments serves as a motivation for each article that follows.

This paper presents initial work on a context-dependent driver distraction detection algorithm called AttenD2.0, which extends the original AttenD algorithm with elements from the Minimum Required Attention (MiRA) theory. Central to the original AttenD algorithm is a time buffer which keeps track of how often and for how long the driver looks away from the forward roadway. When the driver looks away the buffer is depleted and when looking back the buffer fills up. If the buffer runs empty the driver is classified as distracted. AttenD2.0 extends this concept by adding multiple buffers, thus integrating situation dependence and visual time-sharing behaviour in a transparent manner. Also, the increment and decrement of the buffers are now controlled by both static requirements (e.g. the presence of an on-ramp increases the need to monitor the sides and the mirrors) as well as dynamic requirements (e.g., reduced speed lowers the need to monitor the speedometer). The algorithm description is generic, but a real-time implementation with concrete values for different parameters is showcased in a driving simulator experiment with 16 bus drivers, where AttenD2.0 was used to ensure that drivers are attentive before taking back control after an automated bus stop docking and depot procedure. The scalability of AttenD2.0 relative to available data sources and the level of vehicle automation is demonstrated. Future work includes expanding the concept to real-world environments by automatically integrating situational information from the vehicles environmental sensing and from digital maps.

Objective. The objective of this paper is to present a driver sleepiness detection model based on electrophysiological data and a neural network consisting of convolutional neural networks and a long short-term memory architecture. Approach. The model was developed and evaluated on data from 12 different experiments with 269 drivers and 1187 driving sessions during daytime (low sleepiness condition) and night-time (high sleepiness condition), collected during naturalistic driving conditions on real roads in Sweden or in an advanced moving-base driving simulator. Electrooculographic and electroencephalographic time series data, split up in 16 634 2.5 min data segments was used as input to the deep neural network. This probably constitutes the largest labeled driver sleepiness dataset in the world. The model outputs a binary decision as alert (defined as <= 6 on the Karolinska Sleepiness Scale, KSS) or sleepy (KSS >= 8) or a regression output corresponding to KSS epsilon [1-5, 6, 7, 8, 9]. Main results. The subject-independent mean absolute error (MAE) was 0.78. Binary classification accuracy for the regression model was 82.6% as compared to 82.0% for a model that was trained specifically for the binary classification task. Data from the eyes were more informative than data from the brain. A combined input improved performance for some models, but the gain was very limited. Significance. Improved classification results were achieved with the regression model compared to the classification model. This suggests that the implicit order of the KSS ratings, i.e. the progression from alert to sleepy, provides important information for robust modelling of driver sleepiness, and that class labels should not simply be aggregated into an alert and a sleepy class. Furthermore, the model consistently showed better results than a model trained on manually extracted features based on expert knowledge, indicating that the model can detect sleepiness that is not covered by traditional algorithms.

The objective of this study was to compare the development of sleepiness during manual driving versus level 2 partially automated driving, when driving on a motorway in Sweden. The hypothesis was that partially auto-mated driving will lead to higher levels of fatigue due to underload. Eighty-nine drivers were included in the study using a 2 ? 2 design with the conditions manual versus partially automated driving and daytime (full sleep) versus night-time (sleep deprived). The results showed that night-time driving led to markedly increased levels of sleepiness in terms of subjective sleepiness ratings, blink durations, PERCLOS, pupil diameter and heart rate. Partially automated driving led to slightly higher subjective sleepiness ratings, longer blink durations, decreased pupil diameter, slower heart rate, and higher EEG alpha and theta activity. However, elevated levels of sleepiness mainly arose from the night-time drives when the sleep pressure was high. During daytime, when the drivers were alert, partially automated driving had little or no detrimental effects on driver fatigue. Whether the negative effects of increased sleepiness during partially automated driving can be compensated by the positive effects of lateral and longitudinal driving support needs to be investigated in further studies.

Objective:

To assess the attentional demand of different contextual factors in driving.

Background:

The attentional demand on the driver varies with the situation. One approach for estimating the attentional demand, via spare capacity, is to use visual occlusion.

Method:

Using a 3 × 5 within-subjects design, 33 participants drove in a fixed-base simulator in three scenarios (i.e., urban, rural, and motorway), combined with five fixed occlusion durations (1.0, 1.4, 1.8, 2.2, and 2.6 s). By pressing a microswitch on a finger, the driver initiated each occlusion, which lasted for the same predetermined duration within each trial. Drivers were instructed to occlude their vision as often as possible while still driving safely.

Results:

Stepwise logistic regression per scenario indicated that the occlusion predictors varied with scenario. In the urban environment, infrastructure-related variables had the biggest influence, whereas the distance to oncoming traffic played a major role on the rural road. On the motorway, occlusion duration and time since the last occlusion were the main determinants.

Conclusion:

Spare capacity is dependent on the scenario, selected speed, and individual factors. This is important for developing workload managers, infrastructural design, and aspects related to transfer of control in automated driving.

Application:

Better knowledge of the determinants of spare capacity in the road environment can help improve workload managers, thereby contributing to more efficient and safer interaction with additional tasks.

Even though often travelling on the same roads, it has been shown that cyclists and car drivers interpret their environment differently, which can lead to misunderstandings and collisions. Based on the Minimum Required Attention (MiRA) theory and the Salience, Effort, Expectancy, Value (SEEV) model, it is investigated whether the attentional requirements put on drivers and cyclists are different in urban intersections, and how difficult it is to fulfil the requirements for the two road user groups. Additionally, glance data from 23 participants who both cycled and drove along an urban route are compared with respect to information sampling strategies and the fulfilment of attentional requirements depending on its type for three intersections. Generally, more attentional requirements existed for cyclists, and due to where they occur relative to the infrastructure, in combination with the physical aspects of cycling, they are less likely to be fulfilled. This was also corroborated by the empirical data, which showed that requirements clearly visible from the infrastructural design are fulfilled more often than those that are not. Overall, the theoretical evaluation of the infrastructure was confirmed by the empirical data, such that the proposed method can be used as a starting point for a theoretical, human centred evaluation of traffic infrastructure. (C) 2019 Elsevier Ltd. All rights reserved.

With encounters between trucks and cyclists still being a major safety issue and physical as well as technological improvements far from ubiquitous implementation, training truck drivers in anticipatory driving to improve their interaction with cyclists may be a way forward. After a baseline drive in an urban environment, truck drivers inexperienced with urban driving received a dedicated training on anticipatory driving, followed by another drive along the same route several weeks later. The drivers were also interviewed about their opinion about the training. The drivers behaviour changed from before to after training, resulting in a better speed management in general, and a more intensive monitoring of the cyclists. There were also some improvements with respect to the placement in relation to the cyclist, but this effect was limited mainly because truck drivers performed well already before the training. The observed results correspond well to the opinions and feelings about the training that were reported by the drivers in the interview. Thus, driver training can possibly be one contributor to an increase in safety in urban areas.

The effects of driver sleepiness are often quantified as deteriorated driving performance, increased blink durations and high levels of subjective sleepiness. Driver sleepiness has also been associated with increasing levels of electroencephalogram (EEG) power, especially in the alpha range. The present exploratory study investigated a new measure of driver sleepiness, the EEG fixation-related lambda response. Thirty young male drivers (23.6 +/- 1.7 years old) participated in a driving simulator experiment in which they drove on rural and suburban roads in simulated daylight versus darkness during both the daytime (full sleep) and night-time (sleep deprived). The results show lower lambda responses during night driving and with longer time on task, indicating that sleep deprivation and time on task cause a general decrement in cortical responsiveness to incoming visual stimuli. Levels of subjective sleepiness and line crossings were higher under the same conditions. Furthermore, results of a linear mixed-effects model showed that low lambda responses are associated with high subjective sleepiness and more line crossings. We suggest that the fixation-related lambda response can be used to investigate driving impairment induced by sleep deprivation while driving and that, after further refinement, it may be useful as an objective measure of driver sleepiness.

The prevalence of mobile phone usage in traffic has been studied by road-side counting, naturalistic driving data, surveillance cameras, smartphone logging, and subjective estimates via surveys. Here, we describe a custom-made smartphone logging application along with suggestions on how future such applications should be designed. The developed application logs start and end times of all phone interactions (mobile phone applications, incoming/outgoing phone calls and text messages, audio output, and screen activations). In addition, all movements are automatically classified into transport, cycling, walking, running, or stationary. The capabilities of the approach are demonstrated in a pilot study with 143 participants. Examples of results that can be gained from smartphone logging include prevalence in different transportation modes (here found to be 12% while driving, 4% while cycling, and 7% while walking), which apps are being used (here found to be 19% navigation, 12% talking, 12% social media, and 10% games) and on which road types (rural, urban, highway etc.). Smartphone logging was found to be an insightful complement to the other methods for assessing phone use in traffic, especially since it allows the analyses of which apps are used and where they are used, split into transportation mode and road type, all at a relatively low cost.

The many vehicle crashes that are caused by driver sleepiness each year advocates the development of automated driver sleepiness detection (ADSD) systems. This study proposes an automatic sleepiness classification scheme designed using data from 30 drivers who repeatedly drove in a high-fidelity driving simulator, both in alert and in sleep deprived conditions. Driver sleepiness classification was performed using four separate classifiers: k-nearest neighbours, support vector machines, case-based reasoning, and random forest, where physiological signals and contextual information were used as sleepiness indicators. The subjective Karolinska sleepiness scale (KSS) was used as target value. An extensive evaluation on multiclass and binary classifications was carried out using 10-fold cross-validation and leave-one-out validation. With 10-fold cross-validation, the support vector machine showed better performance than the other classifiers (79% accuracy for multiclass and 93% accuracy for binary classification). The effect of individual differences was also investigated, showing a 10% increase in accuracy when data from the individual being evaluated was included in the training dataset. Overall, the support vector machine was found to be the most stable classifier. The effect of adding contextual information to the physiological features improved the classification accuracy by 4% in multiclass classification and by and 5% in binary classification. (C) 2018 Elsevier Ltd. All rights reserved.

Approximately 20-30% of all road fatalities are related to driver sleepiness. A long-lasting goal in driver state research has therefore been to develop a robust sleepiness detection system. Since the alertness level is reflected in autonomous nervous system activity, it has been suggested that various heart rate variability (HRV) metrics can be used as features for driver sleepiness classification. Since the heart rate is modulated by many different factors, and not just by sleepiness, it is relevant to question the high driver sleepiness classification accuracies that have occasionally been presented in the literature. The main objective of this paper is thus to test how well a sleepiness classification system based on HRV features really is. A unique data set with 86 drivers, obtained while driving on real roads in real traffic, both in alert and sleep deprived conditions, was used to train and test a support vector machine (SVM) classifier. Subjective ratings based on the Karolinska sleepiness scale (KSS) was used as ground truth to divide the data into three classes (alert, somewhat sleepy and severely sleepy). Even though nearly all the 24 investigated HRV metrics showed significant differences between sleepiness levels, the SVM results only reached a mean accuracy of 61 %, with the worst results originating from the severely sleepy cases. In summary, the high classification performance that may arise in studies with high experimental control could not be replicated under realistic driving conditions. Future works should focus on how various confounding factors should be accounted for when using HRV based metrics as input to a driver sleepiness detection system.

The first in-depth study on the use of electrocardiogram and electrooculogram for subject-dependent classification in driver sleepiness/fatigue under realistic driving conditions is presented in this work. Since acquisitions in simulated environments may be misleading for sleepiness assessment, performing studies on road are required. For that purpose, the authors present a database resulting from a field driving study performed in the SleepEye project. Based on previous research, supervised machine learning methods are implemented and applied to 16 heart- and 25 eye-based extracted features, mostly related to heart rate variability and blink events, respectively, in order to study the influence of subject dependency in sleepiness classification, using different classifiers and dealing with imbalanced class distributions. Results showed a significantly worse performance in subject-independent classification: a decrease of similar to 40 and 20% in the detection rate of the sleepy class for two and three classes, respectively. Since physiological signals are the ones that present the most individual characteristics, a subject-independent classification can be even harder to perform. Transfer learning techniques and methods for imbalanced distributions are promising approaches and need further investigation.

PurposeThis study examined if speed reduction effects from animal-vehicle collision (AVC) countermeasures are merely local or do extend to a wider area, and what implications the results have on road planning practice regarding AVCs.MethodsTwenty-five drivers drove repeatedly on a 9-km long road stretch in a high-fidelity driving simulator. The development of vehicle speed in the surrounding of an automatic speed camera, a wildlife warning sign and a radio message, were investigated in a full factorial within-subject experiment. The factors wildlife fence (with/without) and forest (dense/open landscape) were also included.ResultsThe radio warning message had the largest influence on vehicle speed with a speed reduction of 8km/h that lasted beyond 1km and 2km after the implementation. Eighty-eight per cent of the drivers reported being made extra aware of AVC due to the radio message, which was also associated with stress, insecurity and unsafety. The warning sign reduced vehicle speed by 1.5km/h, but speed reductions were not significantly reduced 1km after the implementation. Only 8 % of the drivers felt insecure/unsafe after passing the wildlife warning sign, explaining its limited impact on speed. There were no main effects of the automatic speed camera on vehicle speed at longer distances after implementation.ConclusionsWe recommend that AVC countermeasures should be of various design, occur at various segments along the road, and preferably be adaptive and geo-localized to minimize habituation effects on drivers.

Purpose

To examine eye tracking data and compare visual behavior in individuals with normal hearing (NH) and with moderate hearing loss (HL) during two types of driving conditions: normal driving and driving while performing a secondary task.

Methods

24 participants with HL and 24 with NH were exposed to normal driving and to driving with a secondary task (observation and recall of 4 visually displayed letters). Eye movement behavior was assessed during normal driving by the following performance indicators: number of glances away from the road; mean duration of glances away from the road; maximum duration of glances away from the road; and percentage of time looking at the road. During driving with the secondary task, eye movement data were assessed in terms of number of glances to the secondary task display, mean duration of glances to the secondary task display, and maximum duration of glances to the secondary task display. The secondary task performance was assessed as well, counting the number of correct letters, the number of skipped letters, and the number of correct letters ignoring order.

Results

While driving with the secondary task, drivers with HL looked twice as often in the rear-view mirror than during normal driving and twice as often as drivers with NH regardless of condition. During secondary task, the HL group looked away from the road more frequently but for shorter durations than the NH group. Drivers with HL had fewer correct letters and more skipped letters than drivers with NH.

Conclusions

Differences in visual behavior between drivers with NH and with HL are bound to the driving condition. Driving with a secondary task, drivers with HL spend as much time looking away from the road as drivers with NH, however with more frequent and shorter glances away. Secondary task performance is lower for the HL group, suggesting this group is less willing to perform this task. The results also indicate that drivers with HL use fewer but more focused glances away than drivers with NH, they also perform a visual scan of the surrounding traffic environment before looking away towards the secondary task display.

Driver sleepiness contributes to a considerable proportion of road accidents, and a fit-for-duty test able to measure a drivers sleepiness level might improve traffic safety. The aim of this study was to develop a fit-for-duty test based on eye movement measurements and on the sleep/wake predictor model (SWP, which predicts the sleepiness level) and evaluate the ability to predict severe sleepiness during real road driving. Twenty-four drivers participated in an experimental study which took place partly in the laboratory, where the fit-for-duty data were acquired, and partly on the road, where the drivers sleepiness was assessed. A series of four measurements were conducted over a 24-h period during different stages of sleepiness. Two separate analyses were performed; a variance analysis and a feature selection followed by classification analysis. In the first analysis it was found that the SWP and several eye movement features involving anti-saccades, pro-saccades, smooth pursuit, pupillometry and fixation stability varied significantly with different stages of sleep deprivation. In the second analysis, a feature set was determined based on floating forward selection. The correlation coefficient between a linear combination of the acquired features and subjective sleepiness (Karolinska sleepiness scale, KSS) was found to be R = 0.73 and the correct classification rate of drivers who reached high levels of sleepiness (KSS andgt;= 8) in the subsequent driving session was 82.4% (sensitivity = 80.0%, specificity = 84.2% and AUC = 0.86). Future improvements of a fit-for-duty test should focus on how to account for individual differences and situational/contextual factors in the test, and whether it is possible to maintain high sensitive/specificity with a shorter test that can be used in a real-life environment, e.g. on professional drivers.

We have developed a physiological signal platform where presently phonocardiographic (PCG) and electrocardiographic (ECG) signals can be acquired and on which our signal analysis techniques can be implemented. The platform can also be used to store patient data, to enable comparison over time and invoke distance consultation if necessary. Our studies so far indicate that with our signal analysis techniques of heart sounds we are able to separate normal subject from those with aortic stenosis and mitral insufficiency. Further we are able to identify the third heart sound. The platform is being tested in a primary health care setting.

BACKGROUND AND OBJECTIVE: Patients have reported that they perceive their own heart sounds differently after open cardiac surgery than before the surgery. The present study was designed to investigate whether changes in heart sounds can be quantitatively measured. METHOD: Heart sounds were recorded from 57 patients undergoing coronary artery bypass graft (CABG) surgery and from a control group of 10 subjects. The so-called Hjorth descriptors and the main frequency peak were compared before and after surgery to determine whether the characteristics of the heart sounds had changed. RESULTS: At a group level, the first heart sound was found to be significantly different after CABG surgery. Generally, the heart sounds shifted toward a lower frequency after surgery in the CABG group. No significant changes were found in the control group. CONCLUSIONS: Heart sounds are altered after CABG surgery. The changes are objectively quantifiable and may also be subjectively perceived by the patients.

Objective-To investigate use of signal analysis of heart sounds and murmurs in assessing severity of mitral valve regurgitation (mitral regurgitation [MR]) in dogs with myxomatous mitral valve disease (MMVD).

Animals-77 client-owned dogs.

Procedures-Cardiac sounds were recorded from dogs evaluated by use of auscultatory and echocardiographic classification systems. Signal analysis techniques were developed to extract 7 sound variables (first frequency peak, murmur energy ratio, murmur duration > 200 Hz, sample entropy and first minimum of the auto mutual information function of the murmurs, and energy ratios of the first heart sound [S1] and second heart sound [S2]).

Results-Significant associations were detected between severity of MR and all sound variables, except the energy ratio of S1. An increase in severity of MR resulted in greater contribution of higher frequencies, increased signal irregularity, and decreased energy ratio of S2. The optimal combination of variables for distinguishing dogs with high-intensity murmurs from other dogs was energy ratio of S2 and murmur duration > 200 Hz (sensitivity, 79%; specificity, 71%) by use of the auscultatory classification. By use of the echocardiographic classification, corresponding variables were auto mutual information, first frequency peak, and energy ratio of S2 (sensitivity, 88%; specificity, 82%).

Conclusions and Clinical Relevance-Most of the investigated sound variables were significantly associated with severity of MR, which indicated a powerful diagnostic potential for monitoring MMVD. Signal analysis techniques could be valuable for clinicians when performing risk assessment or determining whether special care and more extensive examinations are required.

We have previously developed a method for localization of the first heart sound (S1) using wavelet denoising and ECG-gated peak-picking. In this study, an additional enhancement step based on cross-correlation and ECG-gated ensemble averaging (EA) is presented. The main objective of the improved method was to localize S1 with very high temporal accuracy in (pseudo-) real time. The performance of S1 detection and localization, with and without EA enhancement, was evaluated on simulated as well as experimental data. The simulation study showed that EA enhancement reduced the localization error considerably and that S1 could be accurately localized at much lower signal-to-noise ratios. The experimental data were taken from ten healthy subjects at rest and during invoked hyper- and hypotension. For this material, the number of correct S1 detections increased from 91% to 98% when using EA enhancement. Improved performance was also demonstrated when EA enhancement was used for continuous tracking of blood pressure changes and for respiration monitoring via the electromechanical activation time. These are two typical applications where accurate localization of S1 is essential for the results.

In aortic valve stenosis (AS), heart murmurs arise as an effect of turbulent blood flow distal to the obstructed valves. With increasing AS severity, the flow becomes more unstable, and the ensuing murmur becomes more complex. We hypothesize that these hemodynamic flow changes can be quantified based on the complexity of the phonocardiographic (PCG) signal. In this study, sample entropy (SampEn) was investigated as a measure of complexity using a dog model. Twenty-seven boxer dogs with various degrees of AS were examined with Doppler echocardiography, and the peak aortic flow velocity (V_max) was used as a reference of AS severity. SampEn correlated to V_max with R = 0.70 using logarithmic regression. In a separate analysis, significant differences were found between physiologic murmurs and murmurs caused by AS (p < 0.05), and the area under a receiver operating characteristic curve was calculated to 0.96. Comparison with previously presented PCG measures for AS assessment showed improved performance when using SampEn, especially for differentiation between physiological murmurs and murmurs caused by mild AS. Studies in patients will be needed to properly assess the technique in humans.

The aim of this thesis work has been to develop signal analysis methods for a computerized cardiac auscultation system, the intelligent stethoscope. In particular, the work focuses on classification and interpretation of features derived from the phonocardiographic (PCG) signal by using advanced signal processing techniques.

The PCG signal is traditionally analyzed and characterized by morphological properties in the time domain, by spectral properties in the frequency domain or by nonstationary properties in a joint time-frequency domain. The main contribution of this thesis has been to introduce nonlinear analysis techniques based on dynamical systems theory to extract more information from the PCG signal. Especially, Takens' delay embedding theorem has been used to reconstruct the underlying system's state space based on the measured PCG signal. This processing step provides a geometrical interpretation of the dynamics of the signal, whose structure can be utilized for both system characterization and classification as well as for signal processing tasks such as detection and prediction. In this thesis, the PCG signal's structure in state space has been exploited in several applications. Change detection based on recurrence time statistics was used in combination with nonlinear prediction to remove obscuring heart sounds from lung sound recordings in healthy test subjects. Sample entropy and mutual information were used to assess the severity of aortic stenosis (AS) as well as mitral insufficiency (MI) in dogs. A large number of, partly nonlinear, features was extracted and used for distinguishing innocent murmurs from murmurs caused by AS or MI in patients with probable valve disease. Finally, novel work related to very accurate localization of the first heart sound by means of ECG-gated ensemble averaging was conducted. In general, the presented nonlinear processing techniques have shown considerably improved results in comparison with other PCG based techniques.

In modern health care, auscultation has found its main role in primary or in home health care, when deciding if special care and more extensive examinations are required. Making a decision based on auscultation is however difficult, why a simple tool able to screen and assess murmurs would be both time- and cost-saving while relieving many patients from needless anxiety. In the emerging field of telemedicine and home care, an intelligent stethoscope with decision support abilities would be of great value.

Mild sclerotic thickening of the aortic valve affects 25% of the population, and the condition causes aortic valve stenosis (AS) in 2% of adults above 65 years. Echocardiography is today the clinical standard for assessing AS. However, a cost effective and uncomplicated technique that can be used for decision support in the primary health care would be of great value. In this study, recorded phonocardiographic signals were analyzed using the first local minimum of the auto mutual information (AMI) function. The AMI method measures the complexity in the sound signal, which is related to the amount of turbulence in the blood flow and thus to the severity of the stenosis. Two previously developed phonocardiographic methods for assessing AS severity were used for comparison, the murmur energy ratio and the sound spectral averaging technique. Twenty-nine patients with suspected AS were examined with Doppler echocardiography. The aortic jet velocity was used as a reference of AS severity, and it was found to correlate with the AMI method, the murmur energy ratio and the sound spectral averaging technique with the correlation coefficient R = 0.82, R = 0.73 and R = 0.76, respectively.

Objective - To investigate whether time-frequency and complexity analyses of heart murmurs can be used to differentiate physiologic murmurs from murmurs caused by aortic stenosis (AS) in Boxers. Animals - 27 Boxers with murmurs. Procedures - Dogs were evaluated via auscultation and echocardiography. Analyses of time-frequency properties (TFPs, ie, maximal murmur frequency and duration of murmur frequency > 200 Hz) and correlation dimension (T2) of murmurs were performed on phonocardiographic sound data. Time-frequency property and T2 analyses of low-intensity murmurs in 16 dogs without AS were performed at 7 weeks and 12 months of age. Additionally, TFP and T2 analyses were performed on data obtained from 11 adult AS-affected dogs with murmurs. Results - In dogs with low-intensity murmurs, TFP or T2 values at 7 weeks and 12 months did not differ significantly. For differentiation of physiologic murmurs from murmurs caused by mild AS, duration of murmur frequency > 200 Hz was useful and the combination assessment of duration of frequency > 200 Hz and T2 of the murmur had a sensitivity of 94% and a specificity of 82%. Maximal murmur frequency did not differentiate dogs with AS from those without AS. Conclusions and Clinical Relevance - Results suggested that assessment of the duration of murmur frequency > 200 Hz can be used to distinguish physiologic heart murmurs from murmurs caused by mild AS in Boxers. Combination of this analysis with T2 analysis may be a useful complementary method for diagnostic assessment of cardiovascular function in dogs.

Research in textile sensors has lead to new ways to measure electrocardiograms (ECG). However, additional disturbances from e.g. muscular noise and high skin-electrode impedances often result in poor signal quality. The paper contains a simple application of canonical correlation analysis (CCA) on multi channel ECG signals recorded with textile electrodes. Using CCA to solve the blind source separation (BSS) problem, we intend to separate the ECG signal from the various noise sources. The method (CCABSS) was compared to simple averaging of the ECG channels and to the independent component analysis method (ICA). A heart beat detector was used to evaluate the signal quality. Results show that the signal was completely lost while simulating various noise in 33%, 17% and 7% of the cases for averaging, ICA and CCA, respectively.

There is a growing interest in nonlinear analysis of respiratory sounds (RS), but little has been done to justify the use of nonlinear tools on such data. The aim of this paper is to investigate the stationarity, linearity and chaotic dynamics of recorded RS. Two independent data sets from 8 + 8 healthy subjects were recorded and investigated. The first set consisted of lung sounds (LS) recorded with an electronic stethoscope and the other of tracheal sounds (TS) recorded with a contact accelerometer. Recurrence plot analysis revealed that both LS and TS are quasistationary, with the parts corresponding to inspiratory and expiratory flow plateaus being stationary. Surrogate data tests could not provide statistically sufficient evidence regarding the nonlinearity of the data. The null hypothesis could not be rejected in 4 out of 32 LS cases and in 15 out of 32 TS cases. However, the Lyapunov spectra, the correlation dimension (D2) and the Kaplan-Yorke dimension (DKY) all indicate chaotic behavior. The Lyapunov analysis showed that the sum of the exponents was negative in all cases and that the largest exponent was found to be positive. The results are partly ambiguous, but provide some evidence of chaotic dynamics of RS, both concerning LS and TS. The results motivate continuous use of nonlinear tools for analysing RS data. © 2005 Elsevier Ltd. All rights reserved.

The 3(rd) heart sound (S3) is normally heard during auscultation of younger individuals, but it is also common in many patients with heart failure. Compared to the 1(st) and 2(nd) heart sounds, S3 has low amplitude and low frequency content, making it hard to detect (both manually for the physician and automatically by a detection algorithm). We present an algorithm based on a recurrence time statistic which is sensitive to changes in a reconstructed state space, particularly for detection of transitions with very low energy. Heart sound signals from ten children were used in this study. Most S3 occurrences were detected (98%), but the amount of false extra detections was rather high (7% of the heart cycles). In conclusion, the method seems capable of detecting S3 with high accuracy and robustness.

The 3rd heart sound (S3) is normally heard during auscultation of younger individuals, but it is also common in many patients with heart failure. Compared to the 1st and 2nd heart sounds, S3 has low amplitude and low frequency content, making it hard to detect (both manually for the physician and automatically by a detection algorithm). We present an algorithm based on a recurrence time statistic which is sensitive to changes in a reconstructed state space, particularly for detection of transitions with very low energy. Heart sound signals from ten children were used in this study. Most S3 occurrences were detected (98 %), but the amount of false extra detections was rather high (7% of the heart cycles). In conclusion, the method seems capable of detecting S3 with high accuracy and robustness.

It is sometimes difficult to differentiate between innocent murmurs and pathological murmurs during auscultation. In these difficult cases, an intelligent stethoscope with decision support abilities would be of great value. In this study, using a dog model, phonocardiographic recordings were obtained from 27 boxer dogs with various degrees of aortic stenosis (AS) severity. As a reference for severity assessment, continuous wave Doppler was used. The data were analyzed with recurrence quantification analysis (RQA) with the aim to find features able to distinguish innocent murmurs from murmurs caused by AS. Four out of eight investigated RQA features showed significant differences between innocent murmurs and pathological murmurs. Using a plain linear discriminant analysis classifier, the best pair of features (recurrence rate and entropy) resulted in a sensitivity of 90% and a specificity of 88%. In conclusion, RQA provide valid features which can be used for differentiation between innocent murmurs and murmurs caused by AS.

Heart murmurs are often the first signs of pathological changes of the heart valves, and they are usually found during auscultation in the primary health care. Distinguishing a pathological murmur from a physiological murmur is however difficult, why an “intelligent stethoscope” with decision support abilities would be of great value. Phonocardiographic signals were acquired from 36 patients with aortic valve stenosis, mitral insufficiency or physiological murmurs, and the data were analyzed with the aim to find a suitable feature subset for automatic classification of heart murmurs. Techniques such as Shannon energy, wavelets, fractal dimensions and recurrence quantification analysis were used to extract 207 features. 157 of these features have not previously been used in heart murmur classification. A multi-domain subset consisting of 14, both old and new, features was derived using Pudil’s sequential floating forward selection (SFFS) method. This subset was compared with several single domain feature sets. Using neural network classification, the selected multi-domain subset gave the best results; 86% correct classifications compared to 68% for the first runner-up. In conclusion, the derived feature set was superior to the comparative sets, and seems rather robust to noisy data.

Phonocardiographic signals contain bioacoustic information reflecting the operation of the heart. Normally there are two heart sounds, and additional sounds indicate disease. If a third heart sound is present it could be a sign of heart failure whereas a murmur indicates defective valves or an orifice in the septal wall. The primary aim of this thesis is to use signal processing tools to improve the diagnostic value of this information. More specifically, three different methods have been developed:

• A nonlinear change detection method has been applied to automatically detect heart sounds. The first and the second heart sounds can be found using recurrence times of the first kind while the third heart sound can be found using recurrence times of the second kind. Most third heart sound occurrences were detected (98 %), but the amount of false extra detections was rather high (7 % of the heart cycles).

• Heart sounds obscure the interpretation of lung sounds. A new method based on nonlinear prediction has been developed to remove this undesired disturbance. High similarity was obtained when comparing actual lung sounds with lung sounds after removal of heart sounds.

• Analysis methods such as Shannon energy, wavelets and recurrence quantification analysis were used to extract information from the phonocardiographic signal. The most prominent features, determined by a feature selection method, were used to create a new feature set for heart murmur classification. The classification result was 86 % when separating patients with aortic stenosis, mitral insufficiency and physiological murmurs.

The derived methods give reasonable results, and they all provide a step forward in the quest for an intelligent stethoscope, a universal phonocardiography tool able to enhance auscultation by improving sound quality, emphasizing abnormal events in the heart cycle and distinguishing different heart murmurs.

In this study, we investigate the beat-to-beat respiratory fluctuations in pulse wave transit time (PTT) and its subcomponents, the cardiac pre-ejection period (PEP) and the vessel transit time (VTT) in ten healthy subjects. The three transit times were found to fluctuate in pace with respiration. When applying a simple breath detecting algorithm, 88% of the breaths seen in a respiration air-flow reference could be detected correctly in PTT. Corresponding numbers for PEP and VTT were 76 and 81%, respectively. The performance during hypo- and hypertension was investigated by invoking blood pressure changes. In these situations, the error rates in breath detection were significantly higher. PTT can be derived from signals already present in most standard monitoring set-ups. The transit time technology thus has prospects to become an interesting alternative for respiration rate monitoring. © International Federation for Medical and Biological Engineering 2006.

Recurrence plots (RP) visualize multi-dimensional state spaces and represent the recurrence of states of a system. Recurrence points can be divided into true recurrence points and false recurrence points (also called sojourn points). We introduce the true recurrence point recurrence plot, TRP, a variant of the traditional RP excluding the sojourn points. This is a cleaned up RP free from recurrence points originating from tangential motion, and hence a more robust representation of unstable periodic orbits. The method is demonstrated with three simple systems, a periodic sine wave, a quasi-periodic torus and the x-component of the chaotic Lorenz system

Heart sounds (HS) obscure the interpretation of lung sounds (LS). This letter presents a new method to detect and remove this undesired disturbance. The HS detection algorithm is based on a recurrence time statistic that is sensitive to changes in a reconstructed state space. Signal segments that are found to contain HS are removed, and the arising missing parts are replaced with predicted LS using a nonlinear prediction scheme. The prediction operates in the reconstructed state space and uses an iterated integrated nearest trajectory algorithm. The HS detection algorithm detects HS with an error rate of 4% false positives and 8% false negatives. The spectral difference between the reconstructed LS signal and an LS signal with removed HS was 0 34 0 25, 0 50 0 33, 0 46 0 35, and 0 94 0 64 dB/Hz in the frequency bands 20–40, 40–70, 70–150, and 150–300 Hz, respectively. The cross-correlation index was found to be 99.7%, indicating excellent similarity between actual LS and predicted LS. Listening tests performed by a skilled physician showed high-quality auditory results.

Heart sounds (HS) obscure the interpretation of lung sounds (LS). This letter presents a new method to detect and remove this undesired disturbance. The HS detection algorithm is based on a recurrence time statistic that is sensitive to changes in a reconstructed state space. Signal segments that are found to contain HS are removed, and the arising missing parts are replaced with predicted LS using a nonlinear prediction scheme. The prediction operates in the reconstructed state space and uses an iterated integrated nearest trajectory algorithm. The HS detection algorithm detects HS with an error rate of 4% false positives and 8% false negatives. The spectral difference between the reconstructed LS signal and an LS signal with removed HS was 0.34/spl plusmn/0.25, 0.50/spl plusmn/0.33, 0.46/spl plusmn/0.35, and 0.94/spl plusmn/0.64 dB/Hz in the frequency bands 20-40, 40-70, 70-150, and 150-300 Hz, respectively. The cross-correlation index was found to be 99.7%, indicating excellent similarity between actual LS and predicted LS. Listening tests performed by a skilled physician showed high-quality auditory results.

Severe blood pressure changes are well known in hemodialysis. Detection and prediction of these are important for the well-being of the patient and for optimizing treatment. New noninvasive methods for this purpose are required. The pulse wave transit time technique is an indirect estimation of blood pressure, and our intention is to investigate whether this technique is applicable for hemodialysis treatment. A measurement setup utilizing lower body negative pressure and isometric contraction was used to simulate dialysis-related blood pressure changes in normal test subjects. Systolic blood pressure levels were compared to different pulse wave transit times, including and excluding the cardiac preejection period. Based on the results of these investigations, a pulse wave transit time technique adapted for dialysis treatment was developed and tried out on patients. To determine systolic blood pressure in the normal group, the total pulse wave transit time was found most suitable (including the cardiac preejection period). Correlation coefficients were r = 0.80 ± 0.06 (mean ± SD) overall and r = 0.81 ± 0.16 and r = 0.09 ± 0.62 for the hypotension and hypertension phases, respectively. When applying the adapted technique in dialysis patients, large blood pressure variations could easily be detected when present. Pulse wave transit time is correlated to systolic blood pressure within the acceptable range for a trend-indicating system. The method's applicability for dialysis treatment requires further studies. The results indicate that large sudden pressure drops, like those seen in sudden hypovolemia, can be detected. © The Japanese Society for Artificial Organs 2005.