14:30
Motion
Chair: Natasha Maurits
14:30
15 mins
|
Detection of Freezing of Gait in patients with Parkinson's disease using a deep learning approach
Irene Heijink, Emilie Klaver, Jeroen van Vugt, Richard van Wezel, Marleen Tjepkema-Cloostermans
Abstract: Introduction: Freezing of gait (FOG) is one of the debilitating symptoms experienced by patients with Parkinson’s Disease, and the most common cause of falls in these patients. FOG is described as feet being glued to the floor. Cueing, external stimuli like bars on the floor, can help to overcome FOG. In order to enhance the user experience of cueing devices and to diminish intrusiveness and habituation to cues, on-demand cueing is desired. In addition, objective FOG detection enables monitoring and objective assessment of therapy. In this research, the performance on the detection of FOG of three deep learning classifiers based on acceleration data is studied.
Methods: Acceleration data from four studies with walking tasks ranging from straight walking, turning and a narrow pathway was combined. All experiments were recorded for video annotation of FOG, the gold standard. We evaluated the performance of three classification models for the detection of FOG: a Convolution Neural Network (CNN), MiniRocket and InceptionTime. Five fold cross-validation was applied to estimate an unbiased model performance. The models were evaluated using a receiver operating characteristic (ROC) curve. For the best model we made a comparison between different sensor combinations of acceleration data from the upper legs, lower legs and feet. The best model in combination with the best sensor selection was evaluated on the hold out set.
Results: Seventy-one participants were included in this study, 8% of the data was labelled as FOG. The highest AUC-ROC was reached for the CNN trained on the acceleration data of the lower legs and feet with an AUC-ROC of 0.72, sensitivity of 73.7% (72.5 - 75.0%) and specificity of 60.8% (60.3 - 61.3%) on the test set. The mean AUC-ROC of MiniRocket was 0.10 smaller than the AUC-ROC of the CNN, and the mean AUC-ROC of InceptionTime was 0.04 smaller than the CNN. The difference in mean AUC-ROC for the sensor combinations was 0.01.
Conclusion: The classification algorithm has potential to be implemented in on-demand cueing devices and home monitoring applications for objective FOG detection. Further research is needed to optimize the model and improve the performance.
|
14:45
15 mins
|
Simplified body models evaluation in whole-body sagittal-plane angular momentum perturbations
Junhao Zhang, Michelle van Mierlo, Edwin van Asseldonk, Heike Vallery, Peter Veltink
Abstract: During stable gaits, the whole-body angular momentum (WBAM) appears to be fairly constant through segment-to-segment cancellation and control of joint moments [1, 2]. Yet due to external perturbations or gait impairments, the WBAM can substantially deviate from these stable patterns [3]. Therefore, there is a potential to quantify the state of balance through the variations in the WBAM. To date, video-based motion capture systems are usually used to calculate the WBAM based on a full body model [4]. Our future goal is to use an optimal small set of IMUs to quantify the state of balance, thus it is desirable to assess WBAM from measurements taken from a subset of the body segments. We therefore first investigated segmental contributions to the WBAM under the condition of pure sagittal-plane WBAM perturbations, which aimed to help us choose segments that should be included in simplified body models. We furthermore evaluated the effectiveness of selected simplified body models against the full body model.
Based on our previous data reported in [4], two simultaneous forces with the same magnitude were applied in opposite directions to the pelvis and shoulder to perturb only the WBAM, while participants walked on a treadmill. The contribution of each body segment was evaluated against the estimated full-body WBAM both in terms of pattern similarity (correlation coefficients) and magnitude. The results showed that pelvis, torso, thighs, shanks and feet contributed to 1.24%, 30.43%, 5.08%, 33.03% and 36.15%, while the upper limbs only contributed to 2.56% in total. Considering optimal small set of IMUs in future applications, torso and/or thighs and shanks (with one IMU at torso and one IMU at each shank) should be included in simplified models, pelvis and feet could be potentially added if we add more IMUs at these segments. Our next step is to evaluate the effectiveness of these simplified models against the full body model by means of correlation coefficients, root mean square errors, and incremental similarity in variable perturbations. Future application is to use these simplified body models to quantify the state of balance during variable gaits and functional daily tasks, based on IMUs.
|
15:00
15 mins
|
There is someone controlling my balance but it is not me
Romain Tisserand, Brandon Rasman, Nina Omerovic, Ryan Peters, Jean-Sebastien Blouin, Patrick Forbes
Abstract: The instability of human standing demands that the brain accurately perceive balancing self-motion and determines whether movements originate from self-generated actions or external disturbances. Here, we examined how humans perceive their motor actions while balancing upright by assessing their perception thresholds to external disturbances during active balance and passive (immobile) standing. We find that the conscious sense of balance can be distorted by the corrective control of upright standing. Healthy participants actively balanced or stood immobile on a robotic balance simulator while whole-body or ankle perturbations were injected into the loop of ongoing control. We used psychometric curves to estimate perception thresholds to the imposed perturbations and balance responses, which evoked cues of self-motion that were above and below the statistics of their naturally occurring corrective balance actions. When standing immobile, participants clearly perceived imposed perturbations. Conversely, when freely balancing, participants often misattributed their own corrective responses as imposed motion because their balance system had detected, integrated, and responded to the perturbation in the absence of conscious perception. Importantly, this only occurred for whole-body perturbations (but not ankle perturbations) that remained below the natural variability of ongoing balancing oscillations since they were encoded ambiguously with the balance-correcting responses. These findings reveal that our balance system operates on its own sensorimotor principles that can interfere with causal attribution of our actions, and that our conscious sense of balance depends critically on the source and statistics of induced and self-generated motion cues. They further hold clinical relevance for aging and populations with certain pathologies (e.g., vestibular and cerebellar patients), where the expected increase in sensory and motor noise could widen the natural variability of standing and cause misattributions of self-motion at perturbation magnitudes that may threaten stability.
|
15:15
15 mins
|
Length changes of the medial patellofemoral ligament during in vivo knee motion: a dynamic evaluation using computed tomography.
Miriam Boot, Sebastiaan van de Groes, Hans Dunning, Esther Tanck, Dennis Janssen
Abstract: Introduction
Medial patellofemoral ligament (MPFL) reconstruction is the primary treatment for patients with recurrent patellofemoral instability. Still, the surgery is associated with high complication rates. Clinical outcomes can be improved by a better understanding of MPFL length changes during knee movement. Yet, a clear understanding of normal patellofemoral biomechanics of the MPFL is lacking. This study aimed to assess length changes of the MPFL from 0 to 90 degrees of flexion.
Methods
A high resolution static and medium resolution dynamic CT scan of both knees were obtained in 97 healthy subjects. Static CT scans were obtained in full extension. Dynamic CT scans were obtained during an active flexion-extension-flexion movement (full extension to 90° flexion) in 10s. Static and dynamic CT data were superimposed using image registration and transformations were interpolated to get 3D knee joint models per angle of flexion. Using the knee models, the MPFL length was measured based on anatomic studies from Schöttle’s point on the femur to three insertion points on the superomedial border of the patella (proximal, central, and distal). The shortest wrapping path around the femoral condyle was selected as the MPFL length. Subsequently, MPFL length changes were assessed per flexion angle and expressed as percentual length changes relative to the length in full extension.
Results
The average MPFL length changed less than 5% during knee movement (range: 0.8 – 20.6%). During the first 20-30° of flexion, the average MPFL length decreased with 2%. Beyond 20-30° of flexion, the elongation pattern depended on the patellar attachment site. For the central patellar attachment, the MPFL length restored to its initial length (0 ± 6%), whereas the length remained decreased for the proximal attachment (-3 ± 6%) and even increased for the distal attachment (+4 ± 7%).
Conclusion
The MPFL has on average a near-isometric behaviour during knee flexion but interindividual differences in elongation patterns are large. These findings suggest that a personalized surgical approach may be desirable for reducing complication rates. Future studies should focus on anatomical causes underlying to these differences in elongation patterns.
|
15:30
15 mins
|
Indoor human movement event monitoring with FMCW radar
Reda El Hail, Dominique Schreurs, Peter Karsmakers
Abstract: Healthcare systems encounter many difficulties due to a shorthand of hospital staff members. For instance, in clinics some patient rooms need constant monitoring which evidently cannot be performed by caregivers. Both wearable and contactless technological solutions have been proposed. Recent works have considered radar sensors since such sensors do not need a patient to wear anything, are robust to different lighting conditions and preserve the patients’ privacy.
In this work a frequency modulated continuous waveform radar along with signal processing and machine learning algorithms is assessed for the purpose of recognizing human movement events. Data was collected from 10 people using two radars at two positions (ceiling, wall), and a camera for annotation purposes. The dataset contains 246 “walk” events, 96 “sit” events, 96 “stand-up” events, 42 “hand movement” events, 108 “lay dawn” events and 108 “get up” events. A radar configuration was carefully chosen based on the desired maximum and resolution of range and velocity. Raw signals captured by the radar were processed using a standard processing pipeline into Doppler-time maps.
To detect low-level movement events from the Doppler-time maps two deep learning model architectures, a Convolutional Neural Network (CNN) and a Convolutional Long-Short Term Memory (ConvLSTM), were assessed and compared to each other. Experiments were carried out to determine appropriate hyper-parameter values such as the time horizon for the Doppler- time maps. It was observed that the additional temporal modeling that ConvLSTM provides did not improve performance. The best CNN model achieved 85.5% average recall in a leave-one- person-out experiment. When inspecting the model misclassification in more detail it turned out that there are primarily two explanations for most misclassifications. The first concerns quick events, like sitting, which take less than a second to complete, and the second involves unusual data from volunteers who were moving more slowly compared to others. Furthermore, the radar mounted on the wall had a slightly higher performance in terms of classification accuracy compared to the radar on the ceiling. Based on these preliminary findings, future research will design new models that can deal better with both short and longer events and focus on relative Doppler speed patterns.
|
15:45
15 mins
|
Comparison of distance-limited walk tests and free gait using wearable sensors
Mariano Bernaldo de Quiros, Inge van den Akker-Scheek, Claudine Lamoth, Natasha Maurits
Abstract: Introduction: Distance-limited walking tests, like the 10 m walking test, are common to assess walking ability in patients with motor problems, such as Parkinson’s Disease (PD). Several studies have compared their reliability to longer tests like the 6-minute test, with mixed results depending on the disease and walking ability [1-3]. Wearable sensor technology that patients can wear during their activities of daily living can help understanding the differences between controlled and free gait performance. In this study we compared distance-limited and free-living gait, using wearable sensors and well-known algorithms to obtain gait characteristics.
Methods: Thirty-six healthy participants (18 < 30 years old, 17 > 60 years old), and nine Parkinson’s Disease patients (four in-clinic, five at-home) were recruited and fitted with Inertial Measurements Units on both wrists, right thigh and right shank. All participants, except the at-home patients, were asked to perform a distance-limited walking test (8 m for healthy participants, 10 m for PD) and then to perform the activities of their choice in a free-living environment, with at least one longer walking period being encouraged. Once the data were retrieved, pre-processed and the activities labelled (using video for healthy and in-clinic participants, and a human activity recognition algorithm for at-home patients) several gait characteristics, such as speed, symmetry, harmonicity and entropy were calculated for the distance-limited and free-living walking periods. These characteristics were then compared using ANOVA tests.
Results: Most gait characteristics were similar for distance-limited and free-living walking. However, entropy, which is a more subtle gait characteristic, did differ between the two walking conditions for all studied groups (p < 0.05 after false discovery rate correction).
Discussion: Although distance-limited walking tests provide a simple and fast method to assess walking ability, they may not be fully representative of unrestricted gait. The study of longer, unsupervised periods of free-living walking by means of wearable sensors and human activity recognition algorithms may allow a more ecologically valid assessment of a patient’s walking ability.
References:
[1] Duncan RP, Combs-Miller SA, McNeely ME, et al. Are the average gait speeds during the 10meter and 6minute walk tests redundant in Parkinson disease?. Gait Posture. 2017;52:178-182. doi:10.1016/j.gaitpost.2016.11.033
[2] Amatachaya S, Naewla S, Srisim K, Arrayawichanon P, Siritaratiwat W. Concurrent validity of the 10-meter walk test as compared with the 6-minute walk test in patients with spinal cord injury at various levels of ability. Spinal Cord. 2014;52(4):333-336. doi:10.1038/sc.2013.171
[3] Forrest GF, Hutchinson K, Lorenz DJ, et al. Are the 10 meter and 6 minute walk tests redundant in patients with spinal cord injury?. PLoS One. 2014;9(5):e94108. Published 2014 May 1. doi:10.1371/journal.pone.0094108
|
|