6th Dutch Bio-Medical Engineering Conference
26 & 27 January 2017, Egmond aan Zee, The Netherlands
10:30   Biomedical Signal Processing I
Chair: Natasha Maurits
15 mins
Carolina Varon, Luca Faes, Dries Testelmans, Bertien Buyse, Sabine Van Huffel
Abstract: It is well-known that sleep apnea affects the respiration and the heart rate (HR), and studies have shown that the cardiorespiratory coupling is also compromised during obstructive sleep apnea (OSA) [1]. Furthermore, the classification of hypopneas is challenging, in particular when only ECG-derived features are used. In this context, this study investigates how different ECG-derived respiratory (EDR) signals resemble the respiratory effort during different types of sleep apnea, and how the amount of information transferred from respiration to HR varies according to the respiratory signal used, namely, real or ECG-derived. ECG and respiratory signals recorded from 10 apnea patients of the sleep laboratory of the University Hospitals Leuven were analysed. All signals were segmented into epochs containing normal activity and different types of respiratory events, namely, obstructive, central or mixed apneas, and obstructive or non-obstructive hypopneas. From the ECG, the tachogram and three different EDR signals were computed. These EDRs were compared against the respiratory effort measured on the thorax and abdomen, using correlation and coherence. Furthermore, the transfer entropy was computed between the tachogram and both the real and estimated respiratory signals using information dynamics [2]. The differences between the entropy estimates of different apnea types and normal segments were evaluated using the Kruskal-Wallis test with 95% of confidence. Since 6 different groups (i.e. 5 apnea types and 1 normal group) were compared all versus all, a multi-comparison test was used with Bonferroni correction. Results show that the values of coherence and correlation between the EDRs and the respiratory effort are significantly higher for normal events than for apnea episodes. Moreover, the information transfer is reduced during all types of apneas/hypopneas, and they indicate that the EDR might not capture all variations in cardiorespiratory dynamics during hypopneas. Finally, the information transfer computed using the real respiratory signal achieve accuracies of up to 85% in the detection of sleep apnea with 76% of hypopneas correctly detected, compared to 79% achieved using the EDR with only 63% correctly identified hypopneas. These results suggest that the use of the EDR might reduce the performance of sleep apnea detectors that are based solely on the ECG.
15 mins
Maarten Heres, B.C.Y, Tchang, T. Schoots, Marcel Rutten, Frans van de Vosse, Richard Lopata
Abstract: Background: Muscle perfusion must increase during exercise to meet the rising demand for oxygen supply and waste product removal. Assessment of perfusion dynamics during exercise can provide diagnostic information in e.g. muscle and endothelial disorders. Power Doppler Ultrasound (PDUS) has been recognized as a sensitive tool for the detection of moving blood volume, a perfusion related parameter [1,2]. However, application during exercise is challenging, due to the dependency of PDUS signal on probe location, insonification angle and tissue movement. In this study we present a method for controlled quantitative measurement of blood volume changes with PDUS on skeletal muscle, before, during, and after exercise. In volunteers studies, we demonstrate its feasibility and reproducibility in combination with standardized exercise protocols. Materials and methods: Power Doppler measurements were performed with a MyLab US system (ESAOTE). A 7.5 MHz linear array was fixated with a custom made probe holder on the rectus femoris or the gastrocnemius muscle. Leg extension was performed for increasing load until exhaustion (N = 12). The calf raise exercise (N = 8) was performed for an increasing number of repetitions (3x, 10x, 20x). Exercise was interrupted by a rest period of 20s to allow PDUS measurements free of movement. Data were processed, including noise filtering and attenuation correction. The total number of PDUS pixels were divided by the total area of interest, as a measure of moving blood volume (MBV). The kinetics of MBV increase and recovery were estimated from these data. To test reproducibility, measurements were repeated on two separate days, two weeks apart. Results: With the probe fixated to the leg by the probe holder, the imaging position was kept constant during the entire measurement. A significant increase of PDUS signal was detected in both leg extension and calf raises. For the calf raise experiment, intra-class correlation (ICC) values of the MBV varied between 0.4 – 0.8, depending on the number of repetitions, stressing the need for a controlled experiment. The alternative, leg extension, also allowed for measuring the slope of blood volume increase. The ICC for measurements on separate days was 0.6-0.7. Reproducibility can be improved overall with activity restriction, since muscle perfusion is highly dependent on activity prior to the measurement. 1. J. M. Rubin et al., “Fractional moving blood volume: estimation with power Doppler US.,” Radiology 197(1), 183–190 (1995) [doi:10.1148/radiology.197.1.7568820]. 2. J. S. Newman, R. Adler, and J. M. Rubin, “Power Doppler Sonography: Use in Measuring Alterations in Muscle Blood Volume after Exercise,” J. Diagnostic Med. Sonogr. 13(5), 266–266 (1997) [doi:10.1177/875647939701300527].
15 mins
Lieven Billiet, Sabine Van Huffel
Abstract: In an expanding biomedical field, ever more data is made available to clinicians due to newly arising modalities or newly defined features on known signals. It allows to draw more powerful conclusions, but also complicates a clinician’s task: he or she might easily be overwhelmed by the amount of information. As a result, many attempts have been made to create clinical decision support systems [1]. Several well-known machine learnings techniques allow for e.g. classification tasks. However, common techniques are mostly black boxes, whereas interpretability is a key requirement for acceptance in a clinical environment. Already since decades ago, medical scoring systems have been used to summarize medical knowledge and serve as decision support. One example is Alvarado for appendicitis [2]. It indicates important variables and references values, attributing points if these have been exceeded. The total score relates to a risk of (in this case) appendicitis. However, such scoring systems are based on experience and rules of thumb rather than objective evidence. We propose a framework for semi-automatic extraction of such systems from measured data. The system we propose, Interval Coded Scoring (ICS) [3], combines a specific data representation with sparse optimization. First, every variable is expanded to binary features corresponding to statistically-derived bins in its range. Every bin gets a separate weight. Finally, sparse total variation optimization techniques allow to both reject uninformative initial variables and to derive a simple model with as few ranges per variable as possible. The approach is semi-automatic in the sense that the user indicates the acceptable trade-off between model simplicity and performance by selecting a regularization parameter . The ICS approach has been successfully tested to detect both main effects and interactions in a synthetic data set. The influence of noise and training set size have been quantified. Moreover the scaling of the execution time with both the set size and the number of variables has been determined. This sensitivity analysis presents evidence of the potential of the system for real biomedical applications. Moreover, its usefulness has been demonstrated on public data sets from the UCI Machine Learning database, with highly accurate results. REFERENCES [1] E. S. Berner Ed., Clinical Decision Support Systems: Theory and Practice, 2nd Edition, Health Informatics Series, Springer, 2007. [2] A. Alvarado, “A practical score for the early diagnosis of acute appendicitis”. Annals of Emergency Medicine, 1986, 15, pp 557-564. [3] L. Billiet, S. Van Huffel and V. Van Belle, “Interval Coded Scoring with Interaction Effects: A Sensitivity Study”, 5th Intl Conf on Pattern Recognition Application and Methods, 2016
15 mins
Marzieh Hamdast, Jens Muehlsteff
Abstract: Optimal adjustment of analgesia (pain relief dosage) during surgery is beneficial for patients, as well as clinicians in terms of patient safety and patient experience. In the case of a nociceptive stimulus being applied with inadequate analgesia, the autonomic nervous system activates and then results in changes in various physiological parameters such as Heart Rate Variability (HRV) and PhotoPlethysmoGraphic pulse wave Amplitude (PPGA). Over the last decade, various techniques have been studied to estimate intraoperative nociception. Two state-of-the-art nociception-anti-nociception indices that have been commercialized as guidance metrics for analgesic level are: The Analgesia Nociception Index (ANI) and the Surgical Stress Index (SSI). However, the ANI method that was used can be corrupted by events such as arrhythmia and very low respiratory rates, while the SSI approach might fail in hypovolemic and hemodynamic conditions. The aim of this study is to investigate SSI behavior using PhotoPlethysmoGraphy (PPG), which is non-invasively measured at the peripheral and central sites from patients during abdominal surgery. The performance of central and peripheral site indices is compared to the HRV-related index (ANI) that was used. To estimate the sensitivity of three site indices at the fingertip, forehead and nose bridge, a list of features are derived to assess changes in index features in response to nociception and antinociception events such as 'first incision', '20% HR increase' and 'analgesic effect'. The change in each event is calculated over a defined time-window size between a baseline period before each event and a response period after. Based on the results of this work, the SSI with regard to centrally PPG measurements responses more robust to nociception and antinociception events than peripherally PPG measurement at the fingertip. The derived forehead and nose bridge SSI show promise in nociception and anti-nociception indications respectively, in the abdominal data set collected during general anesthesia.
15 mins
A.L. van der Veen, O. Martinez-Manzanera, D.A. Sival, N.M. Maurits
Abstract: Ataxia is a disorder characterized by impaired balance and poor movement coordination. These symptoms result from an impairment in the brain areas involved in coordination and in the integration of sensory and motor signals. Impaired coordination can be seen as the major symptom of this disorder, but ataxia can also appear as a symptom in other movement disorders. This distinction determines the difference between two phenotypes: ‘core ataxia’ and ‘mixed ataxia’[1]. However, the correct identification of these phenotypes can be difficult and can be prone to subjective interpretation. To evaluate ataxia, rating scales are used. The Scale for the Assessment and Rating of Ataxia (SARA) is the most widely used scale for the assessment of ataxia. The purpose of this study is twofold. First we aim to develop a system for automatic identification of ‘core ataxia’, ‘mixed ataxia’ and ‘no ataxia’ employing data from inertial sensors and a supervised classifier (random forest). Second, we aim to determine feature importance according to the classification and use these features as feedback for the evaluators to improve the agreement on the phenotypic assessment. Twenty patients with ataxia symptoms and a similar number of age-matched controls executed the tasks described in the Scale for the Assessment and Rating of Ataxia (SARA) while wearing inertial measurements units (IMUs). For gait, six IMUs (each composed of three accelerometers and three gyroscopes) were employed to record movements of the trunk, the lower back and the lower limbs. Three evaluators assessed the videos of the participants and assigned them to one of the possible phenotypes. We will employ the gait segmentation algorithm described by Salarian et al [2] to obtain individual strides and derive temporal and spatial features. These features and the assessments of one evaluator will be used as input for the classifier. We will repeat this procedure employing the assessments of the other evaluators as input. We will use the feature importance of each classification procedure to determine differences in the relevance of each feature. We expect that feature importance feedback will help to obtain consensus in phenotypic evaluations of conditions that are difficult to evaluate and thus improve the correct identification of these conditions.
15 mins
Roel Meiburg, Marcel Rutten, Frans van de Vosse
Abstract: Valvular Heart Disease (VHD) currently affects 2.5% of the population, but is overwhelmingly a disease of the elderly, and consequently on the rise. Untreated VHDs are associated with increased morbidity and mortality, while valve replacement procedures (e.g., TAVI) are expensive, invasive and potentially subject to other adverse effects (e.g., valve thrombosis [1]. Current clinical guidelines are limited to gross clinical parameters such as ejection fraction and blood pressure[2], which are limited in the prediction of post-treatment ventricular function.. Mathematical models can provide additional data that cannot be (feasibly) measured clinically, and predict patient hemodynamics post-operatively, so as to provide decision support [3]. Furthermore, model uncertainty is quantifiable, and therefore provides a good assessment of patient parameters. In this study, we apply a model, and model uncertainty analysis, to the application of Trans-catheter Aortic Valve Implantation, or TAVI. Decision support requires adaptation of model parameters to patient-specific conditions, i.e., initial conditions, parameters and boundary conditions. In the case of TAVI decision support, cardiac-based CFD calculations of the aortic valve are combined with a personalized time-varying elastance and Windkessel model, requiring a total of 24 input parameters. Sensitivity analysis was performed using adaptive sparse generalized Polynomial Chaos Expansion (agPCE), where model response is expanded into a finitie series of Lagrange polynomials that depend on the model input (i.e., a meta-model). Using parameter ranges from literature reduces the number of critical model parameters to five. The five remaining parameters are estimated using an augmented Unscented Kalman Filter(UKF). UKF is an algorithm that estimates system states (in this case flows, pressures and volumes) and corresponding estimate uncertainties, which are subsequently corrected by (noisy) measurements. By augmenting the state vector with the parameters, both states and parameters can be estimated. Finally, agPCE is applied to quantify the model output uncertainty with the newly estimated parameters and their respective uncertainties. The results indicate that good assessments of global circulatory parameters (SV, BP, and resistance and compliance of the systemic circulation are required for the model to give useful results. For practical applications this is good news, since such parameters can be measured in a patient fairly accurately.