11:30   Neurostimulation, Neurophysiology and Monitoring Chair: Natasha Maurits Neural code of reflex to investigate spinal adaptation in humans Utku Sukru Yavuz, Laura Schmid Abstract: The reflex excitability of a muscle is an important metric in clinics and neuroscience to understand adaptation in the neuromuscular system to motor development (e.g., ageing) or motor impairment (e.g., spinal cord injury). However, our knowledge of the neural code of reflex excitability is limited. Because current methods are based on evoked compound electromyography signal (reflex) which cannot be used to identify motoneuron properties contributing to excitability profile. This research aims to understand better how reflex excitability is distributed among a motoneuron population and whether the distribution characteristics can be utilized to determine spinal adaptation. To this end, we analyzed reflex (electrically induced) amplitude variability among a MU population using experimental and simulation data. In-vivo MUs were identified by decomposing high-density surface EMG signals from tibialis anterior muscles. The reflex is elicited by stimulating the common peroneal nerve during sustained dorsiflexion contraction (10% and 20% of maximum voluntary contraction). In the simulation, we created 200 motoneurons using a model with soma and dendrite compartments. To obtain a motoneuron pool, the membrane resistance and compartment size were exponentially distributed across motoneurons. The supraspinal neural drive was modelled with common and independent noise, while the monosynaptic reflex input (EPSP) was kept the same for the entire pool. Both experimental and simulation data showed that the reflex amplitude is nonlinearly influenced by the discharge probability (discharge rate and variability of inter-spike intervals). Overall, these factors build up uncertainty and make single MU reflex amplitude a highly variable predictor to determine change in reflex excitability. However, the uncertainty due to the variation in reflex amplitude can be redeemed by using the probability of a large sample from the entire population. Our results showed that the probabilistic model of a larger population can be a reliable proctor of spinal adaptation. Such direct feedback from spinal neurons would provide objective evidence to understand the severity/level of the motor impairment and to help physiotherapists and physicians design a rational strategy during ongoing therapeutical treatment. Tactile and Electric Neurostimulation in an Earthworm to Develop a Magnetic Field Based Neurostimulation Method Towards the Micro Scale Johan Meyer Abstract: The scope of this work is to scale magnetic stimulation methods, such as transcranial magnetic stimulation, down so the system becomes wearable and eventually implantable [1]. Using magnetic fields for stimulation, the direct metal-tissue interface can be avoided, potentially reducing scarring and damaging chemical reactions might be avoided. First, the technology is tested on a simple animal model, an earth worm (Lumbricus terrestris). This work aims to answer: What are the field parameters required for neural excitation of an earth worm axon when using an electromagnet for neurostimulation? The earth worm has a central nerve cord with two lateral (LGF) and a medial (MGF) giant myelinated axon fibre (50um and 70um diameter respectively). First, the earth worm is sedated in 10% ethanol. Then, two pairs of insect pin electrodes pierce the worm for differential measurement and stimulation with respect to a 5th reference pin. Tactile stimulation is achieved by mechanically tapping the head or tail. The results show tactile stimulation triggers the MGF or LGF depending on the stimulation site. Electrical stimulation triggers both MGF and LGF. The extracellular measured pulse shape of the action potential is a biphasic 1.8ms pulse duration, 80 ± 10uV peak to peak and 2.4ms, 40 ± 10uV peak to peak for the MGF and LGF. The pulse shapes due to electrical and tactile stimulation can, within measurement uncertainty, be considered equal. A voltage versus pulse duration (strength-duration) curve is determined for four worms. The average of these curves is fitted to the Lapicque equation giving a 1.4 ± 0.4V rheobase and a 0.16 ± 0.06ms chronaxie. The electric field in the earth worm due to the voltage on the needles during electric stimulation is estimated at 220 ± 60V/m using a model. By discharging a capacitor into a coil, a 5us half sine pulse, 1.2kA amplitude, can be achieved creating a 4us pulse duration, 2400V/m peak amplitude electric field pulse. This is below the strength-duration curve and is not yet sufficient for repeatable neurostimulation. In future work, the electromagnet stimulation circuit will be altered to provide a stimulus that falls above the strength-duration curve. [1] Lee SW, Fallegger F, Casse BD, Fried SI. Implantable microcoils for intracortical magnetic stimulation. Sci Adv. 2016 Personalized Stress Forecasting Utilizing Multivariate Sparse Data Xueyi Wang, Claudine Lamoth, Elisabeth Wilhelm Abstract: Objective/Introduction: Mental health disorders, including depression and anxiety, impose a significant economic burden on the global economy, costing approximately 1 trillion annually. Among various lifestyle factors, stress is a pervasive issue in contemporary society, exerting significant adverse effects on both mental and physical health, productivity, and overall quality of life [1]. Chronic stress is associated with numerous health complications, including cardiovascular diseases, obesity, diabetes, and mental disorders such as anxiety and depression [2]. Consequently, there is an urgent need for effective stress management strategies and interventions that can aid individuals in coping with stress and enhancing their overall well-being. Datasets and Methods: This study introduces a protocol for stress forecasting utilizing sparse data from commercially available Garmin smartwatches. We present a dataset comprising over 1,200 days of smartwatch data collected from 17 subjects. To tackle the challenges posed by domain shifts, we developed adaptive models with convolutional recurrent neural network that effectively process temporal and spatial information, enabling accurate stress prediction. Our adaptive model, which incorporates domain adaptation techniques, exhibits superior performance compared to other base models. The robustness and effectiveness of the protocol in handling sparse data from wearable devices were rigorously validated using a leave-one-subject-out approach. Various testing of input window size (5, 7, 9, 11, 13 days) and prediction window size (1, 3, 5 7, 9 days) combinations revealed optimal values for accurate stress forecasting. Results: The model demonstrates exceptional performance in stress forecasting when utilizing a one-day prediction window and a seven-day training window, while maintaining reliable performance over extended prediction windows. Our models achieve a Root Mean Square Error (RMSE) of 0.196, which is lower compared to the base models of LSTM and CNN, which achieve 0.280 and 0.310, respectively. These findings prove the potential of wearable technology for non-invasive and real-time health management. Conclusions: This work establishes a solid foundation for future research aimed at enhancing the accuracy and responsiveness of stress forecasting using wearable devices. The dataset and adaptive models presented in this study serve as valuable resources for researchers and practitioners in the fields of mental health and wearable technology. Our protocol and findings have far-reaching implications, paving the way for the development of personalized stress management interventions and early warning systems for mental health disorders. [1] S. Cohen, D. Janicki-Deverts, and G. E. Miller, “Psychological stress and disease,” Jama, vol. 298, no. 14, pp. 1685–1687, 2007. [2] M. Kivimaki and A. Steptoe, “Effects of stress on the development and progression of cardiovascular disease,” Nature Reviews Cardiology, vol. 15, no. 4, pp. 215–229, 2018.windows. These findings underscore the the potential of wearable technology for noninvasive and real-time health management. Moreover, our results confirm the adaptability of our model to senior populations and its practicality for individuals requiring continuous stress monitoring. This work establishes a solid foundation for future research aimed at enhancing the accuracy and responsiveness of stress forecasting using wearable devices. The dataset and adaptive models presented in this study serve as valuable resources for researchers and practitioners in the fields of mental health and wearable technology. Our protocol and findings have far-reaching implications, paving the way for the development of personalized stress management interventions and early warning systems for mental health disorders. A Generative Foundation Model for Five-Class Sleep Staging with Arbitrary Sensor Input Hans van Gorp, Merel van Gilst, Pedro Fonseca, Fokke van Meulen, Hans van Dijk, Sebastiaan Overeem, Ruud van Slound Abstract: Gold-standard sleep scoring is based on epoch-based assignment of sleep stages based on a combination of EEG, EOG and EMG signals. However, a polysomnographic recording consists of many other signals that could be used for sleep staging, including cardio-respiratory modalities. Leveraging this signal variety would offer important advantages, for example increasing reliability, resilience to signal loss, and application to long-term non-obtrusive recordings. We developed a deep generative foundation model for automatic sleep staging from a plurality of sensors and any -arbitrary- combination thereof. We trained a score-based diffusion model using a dataset of 1947 expert-labelled overnight recordings with 36 different signals, and achieved zero-shot inference on any sensor set by leveraging a novel Bayesian factorization of the score function across the sensors. On single-channel EEG, the model reaches the performance limit in terms of polysomnography inter-rater agreement (5- class accuracy 85.6%, Cohen’s kappa 0.791). Moreover, the method offers full flexibility to use any sensor set, for example finger photoplethysmography, nasal flow and thoracic respiratory movements, (5-class accuracy 79.0%, Cohen’s kappa of 0.697), or even derivations very unconventional for sleep staging, such as tibialis and sternocleidomastoid EMG (5-class accuracy 71.0%, kappa 0.575). Additionally, we propose a novel interpretability metric in terms of information gain per sensor and show this is linearly correlated with classification performance. Finally, our foundation model allows for post- hoc addition of entirely new sensor modalities by merely training a score estimator on the novel input instead of having to retrain from scratch on all inputs. Unobtrusive Nighttime Movement Monitoring to Support Nursing Home Continence Care by Using Accelerometer Sensors and an XGBoost algorithm Hannelore Strauven, Chunzhuo Wang, Hans Hallez, Vero Vanden Abeele, Bart Vanrumste Abstract: Background: Incontinence is a disorder that can and should be treated and may not be seen as an inevitable effect of ageing [1]. The rising prevalence of urinary incontinence (UI) among older adults, particularly those living in nursing homes (NHs), underscores the need for innovative continence care solutions [2,3]. The implementation of an unobtrusive sensor system may support nighttime monitoring of nursing home residents' movements and, more specifically, detect agitation that is possibly associated with voiding events [4]. Objective: This study explored the application of an unobtrusive sensor system, integrated into a care bed with accelerometer sensors connected to a pressure redistributing care mattress, to monitor nighttime a person's nighttime movement. Methods: Six adult participants followed a seven-step protocol. The obtained dataset was segmented into 20s windows with a 50% overlap. Each window was labelled with one of the four chosen activity classes: in bed, agitation, turn and out of bed. From the Time Series Feature Extraction Library (TSFEL) [5], a total of 1416 features were selected and analysed with an XGBoost algorithm [6]. At last, the model was validated using ‘leave one subject out cross-validation’ (LOSOCV). Results: In total, the dataset in this study encompasses 898 windows. The trained model was able to successfully detect the specified nighttime activities with an overall F1-score of 79.56% and, more specifically, an F1-score of 79.67% for the class 'Agitation'. Conclusions: The results from this study provide promising insights in unobtrusive nighttime movement monitoring and are accepted for publication at JMIR Nursing (doi.org/10.2196/58094). The study underscores the potential to enhance the quality of care for nursing home residents, via a machine learning model based on data from accelerometer sensors connected to a viscoelastic care mattress. Future work: A follow-up study in a NH setting will be conducted in the Erasmus+ project PROCON (Project ID: 101185699) with the objective to foster innovation in NHs, specifically in the promotion of continence in NH residents. Incorporating the residents and monitoring their nighttime behaviour will present new challenges, including limited bed mobility or the need for transfers when immobile. References [1] Miranda A. et al. Psychosocial and societal burden of incontinence in the aged population: a review. Archives of Gynecology and Obstetrics, 277(4):285–290, April 2008. ISSN 0932-0067, 1432-0711. DOI: 10.1007/s00404-007-0505-3. URL http://link.springer.com/10.1007/s00404-007-0505-3. [2] Omotunde, Muyibat and Wagg, Adrian. Technological Solutions for Urinary Continence Care Delivery for Older Adults: A Scoping Review. J Wound Ostomy Continence Nurs. 50(3):p 227-234, May/June 2023. DOI: 10.1097/WON.0000000000000965 [3] World Health Organization. Integrated care for older people: guidelines on community-level interventions to manage declines in intrinsic capacity. World Health Organization, Geneva, 2017. ISBN 978-92-4-155010-9. URL https://iris.who.int/handle/10665/258981. Section: ix, 46 p. [4] Jiaqi Gong et al. Home wireless sensing system for monitoring nighttime agitation and incontinence in patients with Alzheimer’s disease. In Proceedings of the Conference on Wireless Health, WH ’15, pages 5:1–5:8, New York, NY, USA, October 2015. ACM. ISBN 978-1-4503-3851-6. doi: 10.1145/2811780.2822324. URL http://doi.acm.org/10.1145/2811780.2822324. event-place: Bethesda, Maryland. [5] Marília Barandas et al.. TSFEL: Time Series Feature Extraction Library. SoftwareX, 11:100456, January 2020. ISSN 2352-7110. DOI: 10.1016/j.softx.2020. 100456. URL https://www.sciencedirect.com/science/article/pii/S2352711020300017. [6] Tianqi Chen and Carlos Guestrin. XGBoost: A Scalable Tree Boosting System. In Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, pages 785–794, San Francisco California USA, August 2016. ACM. ISBN 978-1-4503-4232-2. DOI: 10.1145/2939672.2939785. URL https://dl.acm.org/doi/10.1145/2939672.2939785. Personalized alpha-motoneuron pool models: towards customized neurorehabilitation Rafael Ornelas Kobayashi, Massimo Sartori Abstract: Human movement arises from the interaction between nervous and musculoskeletal systems, with α-motoneurons (α-MNs) playing a central role. These integrate synaptic inputs and generate the neural drive controlling skeletal muscle activation. The dynamics of α-MN pools varies across individuals and adapt to exercise, aging or injury, influencing motor control. Therefore, measuring the activity of human α-MN pools in vivo is crucial for assessing person-specific neuronal dynamics, neuromuscular adaptations and, in case of lesion, for developing customized motor-restoring interventions. Non-invasive technologies for interfacing with human α-MNs, e.g., surface high-density electromyography (HD-EMG) decomposition, provide a clinically feasible solution to assess the activity of α-MN subsets. Nevertheless, due to intrinsic physiological and technological limitations (e.g., spatial filtering, action potential superposition, etc.), access to the complete α-MN pool remains a challenge. This work combines HD-EMG decomposition, biophysical MN modelling and metaheuristic optimization for creating person- and muscle-specific in silico models of complete α-MN pools. On the experimental side, we recorded HD-EMG from the tibialis anterior (TA) muscle of five healthy and one spinal cord injury subject performing isometric ankle dorsi-flexion in a broad variety of conditions (i.e., levels of force and rates of force development). Subsequently, we used convolution kernel compensation for decoding the in vivo activity of α-MN subsets. On the computational side, we performed a comprehensive sensitivity analysis of the α-MN model parameters to identify the parameters capturing in vivo decoded α-MN dynamics. Thereafter, we estimated surrogates of the synaptic input driving in vivo decoded α-MNs and used it to drive the in silico α-MN models. Through metaheuristic optimization, in silico α-MN models were then calibrated to match in vivo α-MN activity, effectively creating digital copies. Moreover, the statistical distributions of the resulting parameters were used to interpolate the properties of non-decoded in vivo α-MNs (i.e., not identifiable via HD-EMG decomposition) and create person-specific models of the complete α-MN pool innervating TA. This approach provides model-based estimates of complete α-MN pool activity and underlying physiological properties, overall representing a significant advancement for motor science and neurorehabilitation and paving the way for personalized spinal circuitry models to inform motor-restoring interventions.

end %-->