ebook img

All submitted abstracts can also be found here PDF

441 Pages·2017·10.33 MB·English
by  
Save to my drive
Quick download
Download
Most books are stored in the elastic cloud where traffic is expensive. For this reason, we have a limit on daily download.

Preview All submitted abstracts can also be found here

O1 Predicting Gross Motor Function Classification System Level From Patient-Reported Functional Abilities Michael Schwartz1,2, Meghan Munger1 1Gillette Children's Specialty Healthcare, St. Paul, Minnesota, USA, 2University of Minnesota, Minneapolis, Minnesota, USA Introduction: The Gross Motor Function Classification System (GMFCS) is ubiquitous in cerebral palsy (CP) clinical practice and research. The classification is compact, requiring only five cartoons and short descriptions [1]. Traditionally, the GMFCS level is assigned by a health-care provider, though the extended and revised (E&R) version allows for parent rating. The GMFCS has “a particular emphasis on sitting, walking, and wheeled mobility”. We were interested in finding out how the GMFCS was related to a more comprehensive interpretation of “function”, such as the patient/parent-reported Gillette Functional Assessment Questionnaire (FAQ) [2]. Research Questions: Can the GMFCS be determined from the ratings of walking and 22 higher- level functional skills contained in the FAQ? How well do the GMFCS level descriptions reflect functional ability across a wide range of activities? Methods: Our database was queried to find all patient visits with both a therapist-assigned GMFCS rating (I-IV) and patient/parent-report FAQ scores. The data were divided into a training set (75%) and an independent test set (25%). The random forest was used to predict GMFCS based on patient/parent rating of overall walking ability and 22 higher-level functional skills in the FAQ [3]. Results: The full data set consisted of Predicted 6620 observations, and the test set 1614 observations. Accuracy of the model on I II III IV TPR the test set was 90% [Figure 1]. The true positive rate (TPR) ranged from 89% to I 432 38 1 0 92% 92%, and positive predictive value (PPV) II 32 504 24 0 90% ranged from 86% to 93%. The rate of misclassification by more than one level d was exceptionally low (0.3%). e III 1 39 407 12 89% v r e s IV 0 3 11 110 89% Many self-reported functional abilities b O from the FAQ did not conform to the PPV 93% 86% 92% 90% 1614 GMFCS descriptions. As an example, consider running. According to the description of level II “Children have only Figure 1. Performance of the classifier minimal ability to perform gross motor skills such as running and jumping”. However, In our sample, 19% of GMFCS level II children reported running to be “easy”, 40% described it as “a little hard”, while 24% found it “very hard”, and 17% couldn’t run at all. Among GMFCS III children, 10% reported running as “easy”, 21% as “a little hard”, 20% as “very hard”, and 49% couldn’t run at all. There were similar disconnects between other self-reported functional activities (jumping, stair climbing, etc…) and the GMFCS description. Discussion: We have developed an algorithm that can accurately predict GMFCS level from patient/parent-reported functional activity. This model has a variety of uses. In cases where FAQ data are available, a valid and unbiased GMFCS level can be determined based on patient/parent-report. The GMFCS E&R also includes an option for parent report, but still relies on the terse description and cartoons, which may not fully capture the richness of patients’ functional lives. The algorithm can also be useful for retrospective data analysis, since, in some centers, use of the FAQ may have pre-dated use of the GMFCS. The algorithm reveals stark disagreements between patient/parent-reported functional ability and the GMFCS description. This likely arises from the attempt to arbitrarily lump exceptionally diverse individuals into five bins. An intriguing consequence of the algorithm is that it reports both the most suitable GMFCS level, as well as the probability that a patient belongs to each of the four ambulatory GMFCS levels. Thus, an individual may be classified as a [40% 50% 10% 0%] for levels [I II III IV], thus giving a more complete picture of the patient’s overall functional ability. Such a scheme may also pave the way for transition from an ordinal GMFCS to a rational scale version, which would be useful in many research applications. References: [1] “Gross Motor Function Classification System – Expanded and Revised”, CanChild, McMaster University, 2017, Web. 27 Feb 2017. [2] Novacheck, T.F. et al. J Pediatr Orthop (2000) 20: 75. [3] Breiman, L. Machine Learning (2001) 45: 5. O2 Cluster analysis to identify foot motion patterns in children with flexible flatfeet – A statistical approach to detect decompensated pathology Harald Böhm1, Claudia Oestreich2,1, Christel Schäfer1, Chacravarty Dussa1, Leonhard Döderlein1 1Orthopaedic Hospital Aschau, Aschau im Chiemgau, Germany, 2University Giessen, Department of Sport Science, Giessen, Germany Introduction: The pediatric flexible flatfoot (figure) constitutes the major cause of clinic visits for pediatric foot problems [1]. It is a complex deformity in all three planes and its treatment is widely discussed [1]. Hence there is a strong need for classification which may provide indication for treatments. Few approaches have been developed to differentiate subgroups among flexible flatfeet. Bourdet et al.[2] have classified 4 subgroups using static radiographs. Concerning their dynamic function, gait analysis may provide a more reasonable method to discriminate flatfeet into different groups. Research Question: The aim of this study was to classify motion patterns among idiopathic flexible flatfeet using threedimensional foot motion data during walking. Methods: 96 children (192 feet) with clinically diagnosed idiopathic flexible flatfoot between 5 and 17 years were retrospectively included from the database of the gait laboratory. Excluded were syndomal or neuromuscular abnormalities, previous surgeries or tarsal coalition. The Oxford Foot Model (figure) kinematic data was used in order to identify subgroups within flexible flatfeet. Four linear independent angular parameters that were previously shown to be discriminative between flatfeet and typical developing feet were chosen as input for the cluster analysis [3]: 1. peak rearfoot-forefoot dorsiflexion in stance 2.peak rearfoot-tibia eversion in stance, 3. peak rearfoot inversion at push off, 4. mean rearfoot-forefoot abduction in stance. The appropriate number of clusters was evaluated by the hierarchical ward method [4]. The k- means clustering technique was then applied to discriminate patients into the different subgroups [5]. To show the clinical relevance, the number of surgical treatments subsequent to gait analysis was reported for each cluster. Results: Cluster analysis revealed two distinctive flatfoot patterns. Cluster 1 (98 feet) was characterized by achieving hindfoot inversion in late stance and only mild increased peak midfoot dorsiflexion and peak hindfoot eversion. Cluster 2 (94 feet) was characterized by increased dorsiflexion in the midfoot (sagittal), excessive hindfoot eversion (frontal) and forefoot abduction (transversal). Regarding treatment decisions for cluster 1 and 2, 12 of 98 and 57 of 94 feet received arthroereisis, and 4 of 98 and 19 of 94 tarsal osteotomies. Discussion: Cluster analysis discriminated two groups in flatfeet, which appear to be different regarding their severity and number of treatments. Cluster 2 had more severe hindfoot eversion and did not achieve inversion during walking, the midfoot was more bent and the forefoot was considerably abducted. This suggests that feet in cluster 2 can be considered as decompensated and requires treatment.This was confirmed with the higher number of interventions performed in cluster 2. For clinical practice, to distinguish patients between both clusters, logistic regression was applied to all patients. This revealed that peak rearfoot inversion at push-off was the most important discriminator that with every degree of rearfoot inversion, the probability diminishes by 80 % to be part of the decompensated patient cluster 2. References: [1] Bauer et al. J Pediatr Orthop 2016;36:865-9 [2] Bourdet et al. Orthop Trauma Surg Res. 2013;99(1):80-7. [3] Hösl et al. Gait Posture 39 (2014) 23–8. [4] Ward J Am Stat Assoc, (1963) 58, 236–44. [5] Rozumalski et al. Gait & Posture (2009) 155-60. O3 Ground reaction force measurements for gait classification tasks: Effects of different PCA- based representations Djordje Slijepcevic1,3, Brian Horsak1, Caterine Schwab1, Anna-Maria Gorgas1, Michael Schüller2, Arnold Baca2, Christian Breiteneder3, Matthias Zeppelzauer1,3 1St. Pölten University of Applied Sciences, St. Pölten, Austria, 2University of Vienna, Vienna, Austria, 3Vienna University of Technology, Vienna, Austria Introduction: Representations of gait measurements based on principal component analysis (PCA) are among the most successful parametrization techniques in gait classification tasks [e.g. 1, 2]. Compared to solely using discrete parameters, PCA offers a more holistic approach. However, there is no standard process for the computation of such representations so far. In fact, scientific literature differs fundamentally in the arrangement of the employed data and the PCA decomposition method (e.g. linear vs. kernel). Therefore, the aim of this study is to compare different approaches for the derivation of PCA-based representations of ground reaction forces (GRF) and to evaluate their strengths and weaknesses for gait classification. Research Question: What is a best practice for the computation of PCA-based ground reaction force representations for human gait classification? Methods: Gait analysis data from a clinical database were used retrospectively. The database comprises GRF measurements from 279 patients with gait disorders (GD) and data from 161 healthy controls (H), both of various physical composition and gender. Patients were manually classified into four categories, namely a calcaneus (n = 82), ankle (n = 62), knee (n = 69), and hip (n = 66) category. These categories include patients after joint replacement surgery, fractures, ligament ruptures, or related disorders associated with the above-mentioned areas. Bilateral GRF and center of pressure (COP) data were recorded at self-selected walking speed using two force plates. Data were sampled at 2000 Hz and time-normalized to 100% stance. Feature extraction was performed on four different arrangements of the data: (a) PCA of the three components of the GRF of the affected limb (GRFa); (b) PCA of the GRFa with additional COP signals (GRFa+COPa); (c) PCA of the combined GRF for the affected and unaffected limb (GRFa+GRFu); and (d) kernel PCA (KPCA, polynomial kernel) of the GRFa. For this purpose, three data preprocessing types were employed: (i) concatenation of the raw waveforms (e.g. in right Figure) before they were input to a PCA (early fusion), (ii) multiple PCAs of the single waveforms, with concatenation of the resulting principal components (late fusion), and (iii) PCA of commonly used discrete parameters. Resulting principle components retaining 98% of the total variance were then used to train a support vector machine (SVM) for classification of the H vs. GD and H vs. all categories. For the SVM a radial basis function kernel (RBF) and hyper- parameter selection via a grid search were employed. The SVM was trained on a randomly selected data set (65%) and accuracy was evaluated (left Figure) on the remaining data. Results: Discussion: An acceptable level of accuracy was reached for the first classification task. The multivariate task was more difficult to solve due to its complexity, thus resulting in lower accuracy. The KPCA seems slightly advantageous in the multivariate task compared to the traditional PCA of the GRFa, but only when entire waveforms are used. The input of the entire waveforms to the PCA compared to the sole use of discrete parameters generally seems to increase classification accuracy. Late fusion outperformed early fusion, and the use of COP signals improved accuracy, while including the unaffected limb’s GRF had no impact. These data may serve as a guide for future work dealing with GRF for human gait classification. References: [1] Muniz, Gait Posture 2009;29:31-35; [2] Christian, Clin Biomech 2016;33:55-60; O4 Cavovarus foot correction normalizes knee and hip abnormalities in Charcot-Marie-Tooth disease Annika Wallroth, Britta Krautwurst, Nicholas Beckmann, Sebastian Wolf, Thomas Dreher University Hospital, Heidelberg, Germany Introduction: Cavovarus foot deformity (CFD) is typical in patients suffering from Charcot-Marie-Tooth disease (CMT), the most frequent hereditary neurological disorder. Selective muscular atrophy of the foot und shank is followed by progressive bony deformity, flexible in the beginning may result in severe fixed pes cavovarus. Previous research showed that kinematics of the foot significantly improve after operative treatment of the CFD (Dreher et al., 2014) but also need to be applied thoughtfully regarding to the underlying biomechanical conditions (Beckmann et al., 2015). Gait deviations in patients with CFD may also be associated with alterations in transverse plane kinetics and kinematics (Newman et al., 2007, Ferrarin et al., 2012), in terms of altered knee and hip rotation moments and supposed instability of the knee and patella. Research Question: Does combined operative bony and soft-tissue correction of the CFD have an impact on kinetics and kinematics of the ankle, knee and hip? Methods: We examined 24 patients with CMT and bilateral CFD before and after foot reconstruction surgery including a standardized protocol of bony and soft-tissue procedures. In all cases both conventional plug-in-gait analysis and the Heidelberg Foot Measurement Method were used to determine the sagittal, frontal and transverse plane kinematics and kinetics. Comparisons were done using descriptive statistics and linear mixed models to analyze the postoperative change. A p-value of 0.05 was used as cutoff for significance. Results: 3D gait analysis revealed significant reduction in hip abduction and external rotation of the hip and ankle during the stance phase after surgery. These changes were accompanied by significant alterations in the hip kinetics. Transverse plane hip rotation moment normalized significantly. Frontal plane hip and ankle abduction moments normalized postoperatively within the standard deviation of the mean normal reference. Discussion: CFD has an impact on the biomechanical integrity of the hip. Operative procedures of the CFD not only improve foot kinematics but also have a significant impact on the kinematics and kinetics of the knee and hip. External rotation of the hip and its kinetics normalized signifcantly. Frontal plane hip and ankle kinetics improved. This investigation shows the multi-level and multi-plane impact of bony and soft tissue correction on gait in patients with CFD in CMT. As long as those secondary abnormalities are not fixed they can be expected to vanish after foot reconstruction. References: 1. Beckmann, J Foot Ankle Research 2015; 8(65):1-7 2. Dreher, J Bone Joint Surg Am 2014; 96:456-462 3. Ferrarin, Gait & Posture 2012; 35(1):131-137 4. Newman, Gait & Posture 2007; 26:120-127 O5 Perturbation treadmill training improves clinical rating of the motor symptoms gait and postural stability, and sensor-based gait parameters in Parkinson’s disease Heiko Gassner1, Simon Steib2, Sarah Klamroth2, Cristian Pasluosta3,4, Werner Adler5, Bjoern Eskofier3, Klaus Pfeifer2, Jürgen Winkler1, Jochen Klucken1 1Molecular Neurology, University Hospital Erlangen, Friedrich-Alexander University (FAU) Erlangen-Nürnberg, Erlangen, Germany, 2Institute of Sport Science and Sport (ISS), FAU Erlangen-Nürnberg, Erlangen, Germany, 3Pattern Recognition Lab, FAU Erlangen-Nürnberg, Erlangen, Germany, 4Laboratory for Biomedical Microtechnology, Department of Microsystems Engineering, University of Freiburg, Freiburg, Germany, 5Department of Medical Informatics, Biometry and Epidemiology, FAU Erlangen-Nürnberg, Erlangen, Germany Introduction: Impaired gait and postural stability are cardinal motor symptoms in Parkinson’s disease (PD), substantially increase risk of falling, and reduce quality of life [1]. Treadmill training improves gait speed and stride length in PD [2], and might support gait and balance rehabilitation in PD by improving dynamic postural control. In the present study, we investigate a novel form of treadmill training by adding perturbative stimulations to the treadmill surface during walking [3]. Research Question: The aim of this registered, single-blind randomized controlled trial (ClinicalTrials.gov: NCT01856244) was to investigate the effect of an 8 week perturbed treadmill training on the subjective clinical evaluation of the motor symptoms “gait” and “postural stability” in PD. In addition, it was evaluated whether objective spatio-temporal gait parameters recorded by inertial sensor-based gait analysis are able to complement clinical information. Additional functional gait and balance outcomes are presented in a separate publication. Methods and patient characteristics: 43 PD patients were included and randomly allocated to either the experimental group (perturbation treadmill training, PTT, n=21) or a control group (conventional treadmill training, CTT, n=22). Perturbation during treadmill walking was induced by a prototypical treadmill device consisting of a standard medical treadmill (mercury, h/p/cosmos medical GmbH) fixed on a platform with tilt option (zebris Medical GmbH). The platform allows three-dimensional tilting movements induced by three pneumatic actuators with a lifting capacity up to 30 mm. PD patients were secured with a safety harness and performed 16 sessions of 40 minutes treadmill intervention (8 weeks, twice per week). Outcome measures were collected at baseline, after 8 weeks of intervention, and at a 3 months follow-up visit. Clinical motor assessment comprised the Unified Parkinson Disease Rating Scale part III (UPDRS-III), the UPDRS-III subscores “gait” and “postural stability”, and Hoehn and Yahr disease staging (H&Y). Spatio-temporal gait parameters were assessed using a mobile gait analysis system consisting of accelerometers and gyroscopes laterally fixed to shoes [4, 5]. Statistical analysis included within-group effects during intervention (Wilcoxon-Test), and between-group effects of delta values between time points (Mann-Whitney-U-Test). Baseline characteristics including age, gender, height, weight, disease duration, H&Y disease stage, UPDRS-III, medication, and cognition did not significantly differ between groups. Medication was stable and did not differ between groups during the intervention program. Results: After 8 weeks of intervention, motor symptoms rated by UPDRS-III (P=0.001; Effect size Cohen’s d=-1.09), UPDRS-III subscores gait (P=0.023; d=-0.63) and postural stability (P=0.008; d=-0.78), and H&Y disease staging (P=0.001; d=-1.01) significantly improved in the PTT group (within-group comparison), whereas in the CTT group only H&Y disease staging improved (P=0.025; d=-0.55). However, a significant between-group effect of delta values to the benefit of the PTT group in the small cohort was not observed. Swing time variability improved only in the PTT group whereas it worsened in the CTT group (between-group effect: P=0.019; d=-0.58). At 3 months follow-up, within-group effects for UPDRS-III, gait, postural stability, and H&Y persisted only in the PTT group (compared to baseline). Gait parameters were comparable to baseline and did not show differences between groups at the follow-up visit. Discussion: In this study we observed that 8 weeks of perturbed treadmill training improved clinical relevant motor symptoms gait and postural stability which was not present in the group receiving treadmill training without perturbation. Although the clinical score improvement in the PTT group could not reach statistical significance in comparison to CTT in the small cohort, objective, sensor-based gait parameters “swing time variability” improved in PTT only and reached significance between groups. These findings suggest that objective assessment of the regularity of gait is a sensitive parameter complementing clinical assessments for monitoring postural gait control during treadmill therapy in PD. References: [1] Soh et al., Parkinsonism Relat Disord, 2011. 17(1): p. 1-9. [2] Mehrholz et al., Cochrane Database Syst Rev, 2015(9): p. Cd007830. [3] Klamroth et al., Gait Posture, 2016. 50: p. 102-108. [4] Rampp et al., IEEE Trans Biomed Eng, 2015: 62: p. 1089-1097 [5] Klucken et al., PLOS One, 2013. 8(2): p. e56956. O6 Using medical imaging to define the medial-lateral axis of the femur led to significantly different hip rotation kinematics in children with torsional deformities Elyse Passmore1,3, Kerr Graham1,3, Morgan Sangeux1,2 1The Royal Children's Hospital, Melbourne, Australia, 2The Murdoch Children's Research Institute, Melbourne, Australia, 3The University of Melbourne, Melbourne, Australia Introduction: Kinematics and kinetics during gait are used to inform surgical decision making. For example, hip rotation kinematics are considered the key measurement to decide to perform a femoral derotational osteotomy, and to predict surgical outcomes [1]. However, the accuracy and reliability of hip rotation kinematics has been shown to be the least repeatable [2]. The accuracy of hip rotation kinematics depends on the accurate localisation of the medial-lateral axis of the femur. Conventional methods to localise the medial-lateral axis of the femur is through palpation of bony landmarks (the medial and lateral epicondyles), and define the axis from markers, or a Knee Alignment Device (KAD), over the landmarks. Functional calibration methods have also been derived because the medial-lateral axis of the femur determines the sagittal plane flexion-extension axis of the knee joint. Results on the efficacy of functional and conventional methods in the literature varies [3,4]. Furthermore, the methods have only been tested in healthy adults. There is no guarantee that results derived from healthy adult populations translate to clinical populations, especially children with torsional deformities, for which accurate hip rotation kinematics are paramount. Research Question: What is the accuracy of conventional and functional methods to locate the medial-lateral axis of the femur in children with torsional deformities? Methods: Following ethics approval, 20 children (8 with cerebral palsy, 11 with rotational malalignment and 1 with a genetic disorder) with torsional deformities (confirmed with CT medical imaging) and scheduled for 3D gait analysis were recruited to participate in the study. A registered physiotherapist with >5years experience in gait analysis equipped the children with the standard PiG (VICON, UK) marker set, with additional skin markers on the thighs and shanks. The conventional model used the KAD, and the functional methods used the ATT or 2DoFKnee algorithms with flexion-extension and walking calibration movements [3,4]. The posterior-aspect of the condyles were imaged with freehand 3D ultrasound [4] during a static standing trial to determine the location of the medial-lateral axis of the femur. Immediately after walking and without removing the skin markers, the children were transported to the EOS system (a standing low dose bi-plane x-ray system from EOS-imaging, France) to obtain the reference position of the femur and tibia coordinate systems [3], with respect to the skin markers. The main result was the transverse plane angular difference between the various methods and the EOS benchmark. Statistical analysis used general linear statistical model and Tukey’s post hoc tests. To estimate clinical significance, we calculated the proportion of the data more than ±2SD of the variability in hip rotation in our normative dataset from EOS benchmark. Results: Both conventional and functional methods were significantly different from the EOS benchmark. Approximately 50% of the data were greater than +2SD of normal variability from the EOS benchmark. In contrast, freehand 3D ultrasound was only 1° different in average (NS, p=0.146) and all data were within 2SD of normal variability. As a result, hip rotation kinematics from the medical imaging based methods (EOS, freehand ultrasound) were almost identical whereas hip rotation kinematics from conventional and functional methods were significantly different.

Description:
The algorithm can also be useful for retrospective data analysis, since, . Ground reaction force measurements for gait classification tasks: Effects of Results: Repeated measures ANOVA recealed a significant interaction “Measures to Determine Dynamic Balance”, Handbook of Human Motion,
See more

The list of books you might like

Most books are stored in the elastic cloud where traffic is expensive. For this reason, we have a limit on daily download.