11:30   Wearable Chair: Tim Boers Physical activity patterns of patients with chronic low back pain and central sensitization: Insights from a machine learning method Xiaoping Zheng, Michiel Reneman, Bert Otten, Claudine Lamoth Abstract: Introduction: Chronic low back pain (CLBP) is the leading global cause of disability. Central sensitization (CS) is present in a subsample of patients with CLBP. Optimal physical activity (PA) is often recommended in the management of CLBP because it can reduce the risk of disability. However, the evidence of the relationship between PA intensity levels and CLBP is inconsistent, and the knowledge about the association with CS is limited. This study aimed to investigate PA patterns in patients with CLBP and low or high CS using an unsupervised machine learning approach. Methods: Forty-two patients were included (23 CLBP-, a CS Inventory score lower than 40; 19 CLBP+, 40-100). Patients wore a 3D accelerometer for about one week. For each patient, 4 days of data were used for analyses. Accelerometer data were corrected for gravity and the vector magnitude was calculated. For each group, a Hidden semi Markov Model (HSMM) was made to measure the temporal organization and transition of hidden states (PA intensity levels), based on accelerometer vector magnitude. Differences between CLBP- and CLBP+ in duration and occupation of hidden states were assessed with independent t-tests. The transition probability was assessed by Binomial-proportion test. The compositions of corresponding hidden states were assessed with Jensen–Shannon divergence (JSD). Results: The corresponding 5 hidden states of CLBP- and CLBP+ were similar, indicated by JSD. These states were defined as: rest (e.g., sleeping), sedentary (e.g., desk work), light activity (e.g., standing), light locomotion (e.g., slow walking), and moderate-vigorous activities (e.g., fast walking). Significant differences between 2 groups showed that CLBP+ exhibited higher duration and transition probability of active state (light activity, light locomotion, and moderate-vigorous states) and higher duration of inactive state (rest and sedentary states). Discussion: The significant differences in temporal organization and transition of PA levels may suggest that CLBP- and CLBP+ had different PA patterns. CLBP+ group exhibited a prolonged period of activity engagement (overactive) and then had a long period of rest. This PA pattern may suggest that CLBP+ had the distress-endures response pattern. 3D printed graded porous force and strain sensors for wearable sensing applications Nick Willemstein, Ali Sadeghi, Herman van der Kooij Abstract: Wearable sensors are a useful tool for monitoring a wide variety of parameters during motion such as force distribution during walking. Sensors for such applications must be non-intrusive and comfortable to wear. These aspects are essential to minimize the impact on the measurand and foster acceptance by patients. One interesting candidate for this application are soft sensors. These sensors can measure parameters such as pressure, force, and strain while their inherent flexibility allows for comfortable interaction with the user. An example of such soft sensors are porous (foam-like) sensors that can adapt to the user’s body, are lightweight, and can be mechanically programmed (i.e. stiffness can be adjusted). Such porous sensors can be 3D printed through our InFoam printing method [1]. The 3D printing of foam-like structures allows for the fabrication of strain and force sensors in a myriad of geometries, which can be useful for customized sensors. The focus was on investigating the effect of porosity gradient/magnitude on the 1) Stress/strain sensitivity 2) Range (i.e. minimum and maximum) stress/strain 3) Repeatability of the sensor A set of cubes with a carbon black-filled thermoplastic elastomer were printed using our InFoam method with different levels of porosity (gradients). Two experiments were conducted on these samples: 1) Compression tests by a tensile tester for stress/strain behavior 2) Walking test on a force plate with a cube embedded in an insole During all these experiments an Arduino was used to record the electrical resistance of the cube. Subsequently, MATLAB was used to identify and evaluate models. Current results indicate that these porous sensors can be used to estimate the compressive stress and strain quite accurately (>75% fits) yet are still very soft (< 5 megapascals). Whereas the magnitude of stress ranged from a few - >200 kilopascal. In addition, changing the porosity enabled the programming of the strain/stress sensitivity. Thus, soft 3D-printed porous sensors are interesting to provide both stress/strain estimation. In future work, these sensors will need to be tested in a more challenging environment such as with a patient. In addition, multi-modal sensing could be investigated for other loads beside compression. Classification of symptomatic rheumatic heart disease from wearable single-lead ECG signals Amsalu Tomas Chuma Chuma, Carolina Varon, Desalew Mekonnen, Melkamu Hunegnaw, Rik Willems, Bart Vanrumste Abstract: Rheumatic heart disease (RHD) is caused by untreated strep-throat infection from beta-hemolytic group-A streptococci that leads to cardiac valve damages. It mostly affects people at a younger age under 35 years. The health burden of RHD accounts for about one third of the annual cardiac morbidity in the Global South. Echocardiography examination is a gold standard for the diagnosis of RHD. This examination is often impractical in the Global South mainly because of the lack of cardiologists, and the limited echocardiograph machines available at health centers. As a result, RHD is an underdiagnosed disease that leads to an increased morbidity and mortality. One alternative to tackle these limitations, is to develop a more affordable system to detect RHD based on single-lead electrocardiogram (ECG) signals at cardiac wards in the Global South. A limited number of studies have focussed on the use of single-lead ECG for this task. Therefore, this study presents a potential use of single-lead wearable ECG sensors for automatic classification of late-stage RHD using a convolutional neural network (CNN). The CNN feature maps were combined with wavelet coefficients of the input signal to provide distinct spectral ranges of ECG waves. The experimental single-lead ECG dataset was recorded in one of the major cardiac referral hospitals in Ethiopia, which consists of 121 confirmed RHD and 45 normal subjects. Considering common cardiac disease cases in hospitals of the global south, age matched additional 400 subjects’ lead-I rhythms of normal sinus, ischemic, hypertrophic, myocardial infarction and conduction disturbance were added from the Physikalisch-Technische Bundesanstalt (PTB-XL) public dataset. The ratios of each arrhythmia were set to simulate cardiac wards in the global south. From each ECG recording, a slice of 10 seconds was resampled at 250Hz, denoised and then fed to the model. The 5-folds cross-validation performance shows an F1-score of 64.9%, precision of 66.2% and recall of 65%. The results demonstrate that single-lead ECG can be used as a detection tool for symptomatic RHD patients in vulnerable age groups among at-risk communities. This could enhance delivery of point-of-care healthcare in low-resource medical settings. Measuring cerebral oxygenation (NIRS) to monitor orthostatic hypotension Marjolein Klop, Rianne de Heus, Andrea Maier, Carel Meskers, Jurgen Claassen, Richard van Wezel Abstract: Background: Orthostatic hypotension (OH) is highly prevalent in older adults. OH is defined as a blood pressure (BP) drop ≥20 mmHg systolic and/or 10 mmHg diastolic, within 3 minutes after standing. OH is associated with symptoms like dizziness, falls, lower physical and cognitive function, cardiovascular disease and mortality. Currently, OH is diagnosed with single-time-point cuff BP measurements that are not representative for the repeated posture changes that occur in life. These drawbacks may be overcome by use of a non-invasive wearable near-infrared spectroscopy (NIRS) device that uses near-infrared light to measure cerebral oxygenation continuously for longer periods of time. Aim: To validate cerebral oxygenation measured with NIRS as a proxy for BP changes during and after standing up. Methods: Cross-sectional study, including 11 younger (18-35 years) and 30 older adults (≥65 years) with normal and impaired BP responses upon standing. They performed different postural changes, while BP (volume-clamp photoplethysmography) and cerebral oxygenation were measured continuously. Correlations between BP and cerebral oxygenation curves were calculated within participants (based on curve dynamics) and between participants (based on curve characteristics, like maximum drop amplitude and recovery values). Results: Within participants, BP correlated best with oxygenated hemoglobin (O2Hb), but only showed good correlations in the initial 30s after standing up. During baseline and after 30s, all correlations were poor. Characteristics derived from BP and O2Hb measurements correlated poorly, but did show associations in a multilevel linear regression analysis. Conclusion: NIRS-measured O2Hb can capture dynamic BP changes: when BP increased, O2Hb increased as well. However, more complex models are needed for absolute BP estimations, as a larger BP drop did not necessarily result in a larger O2Hb decrease. Individual characteristics might explain these findings. Textile-embedded multi-channel electromyography and musculoskeletal modeling to support clinical decision-making Donatella Simonetti, Bart Koopman, Massimo Sartori Abstract: Introduction: Clinical decision-making requires above all rapidity. Currently, motor deficit evaluation is based on a simplistic and subjective assessment, i.e. a simple 10m walk. It fulfills the main requirement, but it is not accurate, and it is just based on the evaluator's knowledge and experience. Greater accuracy is achieved in biomechanical laboratories where advanced technology together with neuro-musculoskeletal modeling allows to quantify the subject impairment. However, this is made at the expense of rapidity. Our work is aimed to balance clinical rapidity and biomechanical accuracy. We propose to use advanced signal processing techniques and real-time neuro-musculoskeletal modeling integrated into a smart wearable garment. A simple leg sock instrumented with a large-scale multi-electromyography (EMG, 64 channels) grid and inertial sensors (IMUs) allowing to get over the lengthy setup and to prevent human error in the manual electrodes’ placement. The smart clothing together with advanced signal processing techniques provides muscle activation and muscle-tendon unit (MTU) kinematics necessary to finally model the subject-specific musculoskeletal properties. Methods: 8 healthy subjects were equipped with 33 reflective markers and a leg flexible garment instrumented with 64 equally distributed EMG monopolar electrodes. The 64-electrode space is reduced in 5 muscle-specific clusters by applying iteratively the non-negative matrix factorization (NNMF) [1] during slow locomotion at 1km/h. Afterward, 5 average muscle activations were extracted during locomotion at different speeds 1, 3, and 5 km/h, and used as input to an offline EMG-driven musculoskeletal model to estimate ankle torque. Results: The NNMF-based approach was able to locate the muscles and extract averaged activations during each locomotion speed that resembled with good accuracy the activation recorded with bipolar EMG. Afterward, the musculoskeletal model driven by the automatically extracted muscle-specific activation reproduced experimental ankle torques during gait at different speeds. Conclusions: The combination of a soft sensorized garment and the automatic procedure for the extraction of muscle activations added to the framework for neuromuscular modeling has a good potential to become a resource for fast and more accurate clinical decision-making. Acknowledgment This work was founded by EFRO Op Oost GUTs (20913301). The garment is developed in collaboration with TMSi and Bard.zo. Using 2D CNN to detect tonic-clonic seizures based on accelerometer and photoplethysmography signals Chunjiao Dong, Johannes van Dijk, Xi Long, Ronald M. Aarts Abstract: This study aims to design a deep learning model to automatically detect tonic-clonic (TC) seizure events based on accelerometer (ACM) and photoplethysmography (PPG) signals, which are vital signals to illustrate the motion and heart rate changes during TC seizures [1]. Both signals were continuously collected using NightWatch armbands [2], from 44 patients during the night, and each patient was monitored for two to three months.

end %-->