Scientific Paper Audiobooks From SDU's Biomedical Engineering Centre

Probabilistic sensor network model to assess machine learning accuracy outcomes, by Joseph Russell and Jeroen Bergmann, narrated in full by Joseph Russell, as presented at EMBC on 18th July 2020. Full article available at: https://portal.findresearcher.sdu.dk/da/publications/probabilistic-sensor-network-model-to-assess-machine-learning-acc/ This video is part of an initiative by SDU's Biomedical Engineering Centre to improve the accessibility of scientific papers with human-narrated article readings. The text is read out in full in the same manner as a non-fiction unabridged audiobook, with the article displayed page-by-page on-screen for those wishing to see figures, equations and references. Performance of sensor networks can be predicted using probabilistic modelling (PM). This gives an upper bound on the accuracy for a given set of sensors. Incorrect application of machine learning (ML) can be detected when the ML approach yields accuracy levels greater than the upper bound of the PM. This paper demonstrates experimentally how PM can be used for detecting false ML performance outcomes. This probabilistic method can help identify and correct ML issues in datasets obtained from sensor networks such as wearable medical devices. © 2020 by the authors. #biomedicalengineering #academia #scientificwriting #audiobook #probabilisticsensornetwork #sensorfusion #machinelearning

Real-Time Intent Sensing for Assistive Devices with Implications for Minimising Maintenance, by Joseph Russell and Jeroen H.M. Bergmann, narrated in full by Joseph Russell, as published on 6th May 2023 in Prosthesis. Full article available at: https://doi.org/10.3390/prosthesis5020031 This video is part of an initiative by SDU's Biomedical Engineering Centre to improve the accessibility of scientific papers with human-narrated article readings. The text is read out in full in the same manner as a non-fiction unabridged audiobook, with the article displayed page-by-page on-screen for those wishing to see figures, equations and references. Intent sensing is a growing field within medical device control, with major potential applications for assistive devices, such as prosthetics. As many sensors as possible should be utilised to maximise accuracy. The availability of sensors may change over time due to changing surroundings or activities, sensors failing, and electrode contact being lost. The sensor network should be dynamic and modular in nature, continuing to function even when some sensors are unavailable. The management of sensor unavailability may help to reduce the need for device maintenance, particularly in developing nations with limited availability of these services. An algorithm is proposed to classify intent using networked sensors in real time. Data are gathered using human participants wearing four surface electromyography sensors and performing a pseudo-random sequence of grasps. The relationship between time offset and prediction accuracy is investigated, with the algorithm predicting future intent actions up to half a second in advance. Sensor dropout is simulated by randomly replacing sensor readings with recorded noise. The new algorithm is compared to existing algorithms and shown to be more accurate in situations of sensor dropout, with the difference increasing as more sensors become unavailable. This suggests that when reductions in sensing capabilities are likely to occur over time, the modular method is more appropriate for control. © 2023 by the authors.

A five-sensor IMU-based Parkinson's disease patient and control dataset including three activities of daily living, by Joseph Russell, Jemma Inches, Camille B. Carroll and Jeroen H.M. Bergmann, narrated in full by Joseph Russell, as published on 14th August 2023 on Dryad. Full article available at: https://doi.org/10.5061/dryad.fbg79cp1d

This video is part of an initiative by SDU's Biomedical Engineering Centre to improve the accessibility of scientific papers with human-narrated article readings. The text is read out in full in the same manner as a non-fiction unabridged audiobook, with the article displayed page-by-page on-screen for those wishing to see figures, equations and references. Parkinson’s disease is an often-debilitating progressive neurological condition leading to loss of motor control. This dataset contains kinematic sensor data from two groups: one containing 15 patients with Parkinson’s disease, and a control group of 19 participants without any known neurological condition. Participant ages ranged from 40–85, 21 were male, and 13 were female. The participants wore five 9-axis Inertial Measurement Units (IMUs) – one on each upper arm, each lower arm, and on their head. They were asked to perform a calibration pose, followed by three activities: making toast, putting on a cardigan, and unlocking and opening a door, with each activity repeated three times. The IMUs recorded time-series acceleration and orientation data from the moment where the participant was instructed to begin the activity (inception of the idea to act), through to the activity’s completion. This dataset is planned for use in intent-sensing studies for assistive device control but is also applicable for activity recognition. © 2023 by the authors.

A modular, deep learning-based holistic intent sensing system tested with Parkinson’s disease patients and controls, by Joseph Russell, Jemma Inches, Camille B. Carroll and Jeroen H.M. Bergmann, narrated in full by Joseph Russell, as published on 1st November 2023 in Frontiers in Neurology, 14:1260445. Full article available at: https://doi.org/10.3389/fneur.2023.1260445 This video is part of an initiative by SDU's Biomedical Engineering Centre to improve the accessibility of scientific papers with human-narrated article readings. The text is read out in full in the same manner as a non-fiction unabridged audiobook, with the article displayed page-by-page on-screen for those wishing to see figures, equations and references. People living with mobility-limiting conditions such as Parkinson’s disease can struggle to physically complete intended tasks. Intent-sensing technology can measure and even predict these intended tasks, such that assistive technology could help a user to safely complete them. In prior research, algorithmic systems have been proposed, developed and tested for measuring user intent through a Probabilistic Sensor Network, allowing multiple sensors to be dynamically combined in a modular fashion. A time-segmented deep-learning system has also been presented to predict intent continuously. This study combines these principles, and so proposes, develops and tests a novel algorithm for multi-modal intent sensing, combining measurements from IMU sensors with those from a microphone and interpreting the outputs using time-segmented deep learning. It is tested on a new data set consisting of a mix of non-disabled control volunteers and participants with Parkinson’s disease, and used to classify three activities of daily living as quickly and accurately as possible. Results showed intent could be determined with an accuracy of 97.4% within 0.5 s of inception of the idea to act, which subsequently improved monotonically to a maximum of 99.9918% over the course of the activity. This evidence supports the conclusion that intent sensing is viable as a potential input for assistive medical devices. © 2023 by the authors.

BioGuard – Mouthguard-based Sensor Platform, by Joseph Russell, Leonardo De Almeida E Bueno, Runbei Cheng, Yuanzhe Zhao and Jeroen Bergmann, narrated in full by Joseph Russell, as presented on 26th September 2024 at the Inaugural Podium Institute Annual Conference on Sports Medicine & Technology, Oxford. Full article available at: https://portal.findresearcher.sdu.dk/da/publications/bioguard-mouthguard-based-sensor-platform/ This video is part of an initiative by SDU's Biomedical Engineering Centre to improve the accessibility of scientific papers with human-narrated article readings. The text is read out in full in the same manner as a non-fiction unabridged audiobook, with the article displayed page-by-page on-screen for those wishing to see figures, equations and references. The BioGuard is an innovative biometric sensor platform contained within a sports mouthguard. Developed by the Natural Interaction Lab at the University of Oxford, the BioGuard is able to monitor motion, breathing rate, temperature and heart rate from within the user’s mouth. This allows coaches and players to monitor their levels of fatigue, high levels of which have been shown to correlate with not only reduced performance, but also increased risk of injury. The device could therefore be used to inform decisions about rest periods during training and substitution during gameplay, improving player safety. The data could also be used to assess head impacts to predict risk of concussion, to quantify user fitness by estimating VO2Max, to make predictions about player intent, and to quantify coordination between players. To achieve this, several new methodologies were developed. The device itself was designed on a flexible circuit board, with a layout optimised for the available space within the mouth. Protocols for safely encasing the mouthguard within a double-layered EVA sport mouthguard, based on a 3D scan of the user’s mouth, were produced. An algorithm was developed to estimate the volume flow of air inhaled and exhaled, using audio recorded by an on-board microphone. Temperature models were produced relating the recorded temperature at the gum to the core body temperature.Following successful human trials of the individual system components, prototypes of the complete integrated system will shortly undergo usability studies during exercise, with an aim of commercial implementation in the next few years. © 2024 by the authors.

The Effect of Imperfect Labelling on an LSTM Deep Learning Intent Classifier for Assistive Technology, by Joseph Russell and Jeroen Bergmann, narrated in full by Joseph Russell, as published on 25th March 2025 in Palinko, O., et al. Social Robotics. ICSR + AI 2024. Lecture Notes in Computer Science(), vol 15561. Springer, Singapore. Full article available at: https://doi.org/10.1007/978-981-96-3522-1_15 This video is part of an initiative by SDU's Biomedical Engineering Centre to improve the accessibility of scientific papers with human-narrated article readings. The text is read out in full in the same manner as a non-fiction unabridged audiobook, with the article displayed page-by-page on-screen for those wishing to see figures, equations and references. Intent sensing, the estimation of what it is that a human wants to make happen, is a useful input for collaborative robots, allowing them to intuitively detect and anticipate user needs. It is easier to detect an action after it has completed than it is to predict an action in advance. Therefore, it is proposed that a pre-trained after-the-fact classifier could be used to label unsupervised training data for a predictive classifier, and thus allow a self-learning system with a constantly expanding data set, tailored to the specific user. However, the after-the-fact classifier will not be 100% accurate, so this labelling will be imperfect. To determine whether such a system could be viable, it is important to investigate the effect of imperfect labelling on a deep learning intent sensing system. This is tested on time-series data from an existing data set of Parkinson’s disease patients and controls, classified with a Long Short-Term Memory recurrent neural network to estimate intent over time. The results show that while large numbers of label errors cause a negative impact on the classifier accuracy, the classifier is robust to small amounts of label noise, with a label error proportion of 1/12 having only a very small impact (~0.6%) on the accuracy of the classifier. This supports the viability of a self-learning intent sensing system. © 2025 by the authors. Licensee Springer, Singapore. This audio version of the paper was produced with permission from the Springer Nature Editorial Team. The Springer Nature version remains the official version of record.

Method Development for Synchronizing Processor Time of Accelerometer with Computer for Sensor-Equipped Health Measurements, by Marcus Holzschuher, Rebecca Heintz, Joseph Russell, Maria Lißner, Jeroen Bergmann and Anja Winkler, narrated in full by Joseph Russell, as presented at EMBC on 14 July 2025.

Full article available at: https://portal.findresearcher.sdu.dk/en/publications/method-development-for-synchronizing-processor-time-of-accelerome/

This video is part of an initiative by SDU's Biomedical Engineering Centre to improve the accessibility of scientific papers with human-narrated article readings. The text is read out in full in the same manner as a non-fiction unabridged audiobook, with the article displayed page-by-page on-screen for those wishing to see figures, equations and references.

This paper presents a method for the precise synchronization of inertial measurement unit (IMU) data with a computer system clock, using a motor-controlled turntable, to allow more accurate labelling of motion data. Using the case of human reaction time measurement, where motion is recorded as a response to a computer-controlled visual signal, this methodology proved capable of compensating for IMU clock drift and latency between systems, providing a robust and simple solution for applications in a wide range of IMU-sensor-equipped systems.

© 2025 by the authors.

Human Dead Reckoning Using a Particle Filter and Map Constraints, by Joseph Russell and Jeroen H.M. Bergmann, narrated in full by Joseph Russell, as published by Sensors on 2 June 2026, article 3500.

Full article available at: https://doi.org/10.3390/s26113500

This recording is part of an initiative by SDU's Biomedical Engineering Centre to improve the accessibility of scientific papers with human-narrated article readings. The text is read out in full in the same manner as a non-fiction unabridged audiobook, with the article displayed page-by-page on-screen for those wishing to see figures, equations and references (the video component is available on YouTube and Spotify).

This paper presents an approach for tracking a person’s position by integrating inertial measurement unit (IMU) sensor values, utilising a particle filter with known physical constraints, such as a map of the space. While such approaches are well established, the effect of constraint choice in the absence of observation measurements remains poorly understood. The effect of varying the tolerance of these constraints is investigated with data collected from a Movella DOT held by a human participant walking around a 100 m running track. In particular, the dimensions of the map are varied, along with the shape. Results showed (a) that it is viable to correct particle filter error without external sensor feedback, provided constraints are provided, with a mean error of 1.8 m, and (b) there is a minimum acceptable tolerance of map width around the edge of the true activity zone, in this case approximately 8 m, and that tightening the map boundary further than this can counterintuitively lead to reduced accuracy.

The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

© 2026 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license.