Paper Submission & Registration
8th Dutch Bio-Medical Engineering Conference
13:40   Rehabilitation - I
Chair: Peter Veltink
15 mins
Robotic classification of different elbow impairment phenotypes in patients with an upper motor neuron lesion
Levinia van der Velden, Bram Onneweer, Joyce Benner, Claudia Haarman, Marij Roebroeck, Gerard Ribbers, Ruud Selles
Abstract: To improve rehabilitation programs for patients with upper motor neuron lesions, such as stroke and cerebral palsy, quantitative measurements of upper limb impairments can provide valuable diagnostic or prognostic information. With this information, therapy programs and prognosis can personalized by classifying the patient-specific factors of upper limb impairments. In this study, we identified different phenotypes of elbow impairment combinations in 29 stroke patients and 20 patients with cerebral palsy. Four upper limb impairments (muscle weakness, abnormal synergy, spasticity and changes in viscoelastic joint properties) were measured with the Shoulder-Elbow-Perturbator (SEP) for the elbow. K-means cluster-analysis, with the use of Bayesian information criterion, was used to identify different phenotypes. The statistical analysis revealed six statistically distinct subgroups with unique combinations of elbow motor impairments: 1) No upper limb impairments, but increased muscle strength in both flexion and extension, 2) No upper limb impairments in all domains, 3) Reduced extension and flexion strength, 4) Spasticity & abnormal synergy, 5) Spasticity combined with mild extensor muscle weakness, and 6) Changes in viscoelastic joint properties combined with mild spasticity and mild extensor muscle weakness. For clinical interpretation, further research may indicate that group 1 and 2 can be merged together and may need additional screening for coordination or sensibility impairments, while groups 3-6 may need different treatments, such as strength and coordination training, splinting to avoid contractures, or spasticity medication. For further research, additional impairments of multiple joints and/or other than motor impairments need to be measured to optimize the analysis and prognostic value of the phenotyping.
15 mins
Can machine learn from biomechanics? Fatigue detection with machine learning in a fatiguing outdoor run
Luca Marotta, Jasper Reenalda
Abstract: Physical fatigue is a recurrent problem in running, leading to increased risks of getting injured and negatively affecting performance [1]. Identification and management of fatigue helps reducing such negative effects, but is still commonly based on subjective estimates of fatigue. Alternatively, non-invasive sensors used to assess physical fatigue are heart rate monitors and inertial measurement units (IMUs). Heart rate is widely use to assess performance and effort, which are strictly linked to fatigue [2]. IMUs can record biomechanical parameters continuously, which can show changes due to physical fatigue [3][4]. Feeding extensive amounts of biomechanical parameters into machine learning classifiers already led to a fatigue detection accuracy of 90% in working tasks [5]. Yet, few studies have focused on the detection of a fatigue state in running, especially in out-of-the-lab, outdoor environments. Buckley et. al obtained a 75% accuracy in detecting fatigue with a single IMU [6], training on difficult to interpret statistical features. Here we aimed to assess the ability of a machine learning classifier trained on IMU-derived biomechanical features to distinguish between a fatigue and nonfatigue state in an outdoor run, against a classifier based solely on heart rate. We recorded data from 10 participants performing a run on an athletic track with 8 IMUs and a heart rate monitor attached to their body. The run was divided in three consecutive segments: a 4 km run at a comfortable speed; a run until exhaustion (RPE ≥ 16); a 1.2 km run at the same speed of the run pre-exhaustion. We considered the first segment as a ‘non-fatigue’ state and the last segment as a ‘fatigue’ state. We trained a bootstrap-aggregated decision trees machine learning classifier with selected features from the 25-strides moving average of the IMU-derived joint angles and heart rate. We used a leave-one-subject-out cross validation to initially assess the optimal combination of IMU locations to detect fatigue, which resulted in the right upper leg and pelvis. The classifier trained with biomechanical features in the resultant optimal configuration significantly outperforms the classifier trained with heart rate data (Biomechanical Accuracy = 81.21%; Heart Rate Accuracy =65.81 %; p=0.0016).
15 mins
Hip joint torque limits for postural balance of paraplegic subjects with an orthosis activated in the hip joint
Mahboubeh Keyvanara, Mohammad Jafar Sadigh, Kenneth Meijer
Abstract: For the spinal cord injury (SCI) subjects, it is critical to safely maintain postural balance. Different devices are designed to help the postural balance of these subjects. Reciprocal gait orthosis (RGO) are examples of these devices, which, if activated at the hip joint can enable upper body movements too, and this can be very helpful in daily activities of the SCI subjects. Yet, some of the upper body movements can cause instability in postural balance by violating the constraints which guarantee the contact of the feet with the ground. In this research, the torque limits for the movements of the upper body, which will keep the postural balance stable, are determined using three main constraints: friction, center of pressure (CoP) and gravity. If all these constraints are satisfied the orthosis will not slip, neither will it tip over, nor the feet will come off the ground. Hence, the main concentration is understanding the limits for controller of the hip joint which guarantees stable postural balance with no movements of the feet with respect to the ground. The study is done analytically and final results are applied to an existing RGO. The results show, if the friction constraint is satisfied, the gravity constraint will also be satisfied, and hence, the two main constraints which guarantee feet contact with the ground are the CoP and the friction constraints. Also, the final evaluation in this research shows that one parameter which can have effects on all constraints is the shoe used on the orthosis, the length of the shoe affects CoP constraint and the friction between the feet and the ground affects the friction constraint. The results of this study can be helpful in designing devices with active hip joints. Also, the framework presented in the research can be used in other applications such as postural balance of a bipedal robot.
15 mins
Pushing wheelchairs from the side for enhanced communication: Simple mechanical steering compensation
Lucy Bennett, Bram Sterke, Nicole Luitwieler, Heike Vallery
Abstract: When taking care of child wheelchair-users with multiple medical conditions, caregivers require eye-contact in order to communicate and adequately observe the safety of the child. The conventional placement of the caregiver behind the wheelchair hinders this communication. This study aimed to investigate a key step in moving away from conventional pushing positions. The goal was to design a passive wheelchair mechanism to enable direction control when driven eccentrically, i.e. where the caregiver is located next to the wheelchair. An ACCREx-tree framework was used to generate ideas and lead to a patent-pending invention comprising of a castor wheel with modifiable trail and bank angle. This study used a modified wheelchair for testing. The standard front castors of the wheelchair were removed and replaced with a single castor adapted such that both its trail and bank angle could be adjusted. The chair was driven eccentrically through a spring attached at a lateral distance from the centre of mass of the chair. Through motion capture, we measured the turning circle of the wheelchair for different trail and bank angles. It was found that lengthening trail increased the radius of the turning circle. Furthermore, the greater the bank angle of the castor, the more the steering can be biased. Lastly, it was found that driving in a straight line, while pushing eccentrically, could be achieved by selecting a wheelchair-specific bank-angle. These findings show that with modifications to the castors, wheelchairs can be driven eccentrically whilst maintaining a linear trajectory. With further development, this will improve wheelchair interactions for both caretakers and wheelchair-users. The same mechanism could also be used to assist users who use only one hand to propel themselves, for example because of hemiparesis.

end %-->