- Email: [email protected]
An attempt to detect impairment by silhouette-based gait feature
S122
Abstracts / Gait & Posture 39S (2014) S1–S141
simple and easy to interpret measurements for clinical [1]. The aim of the present study was to analyze and compare the gait parameters of PD patients under the influence of dopaminergic medication and/or high frequency DBS using the GDI and the GPS/MAP. Patients/materials and methods: Sixteen patients with PD (eleven male and five female) who were submitted to bilateral high frequency DBS of the STN participated in the study. The gait assessment was conducted using three-dimensional kinematics (SMART-D® BTS) in three conditions: without medication and with stimulation (OFF med/ON DBS); with medication and stimulation (ON med/ON DBS); with medication and without stimulation (ON med/OFF DBS). The label of the markers and the processing of the biomechanical model to obtain kinematic data were performed using Vicon Nexus® software and the Plug in Gait® model. The kinematic data were imported into a spreadsheet, where a mathematical routine was used to calculate the GDI and GPS/MAP in different conditions. The Unified Parkinson’s Disease Rating Scale (UPDRS) part III also was applied during the three conditions. The data were analyzed using the variance for repeated measures test (ANOVA), with the level of statistical significance set at p < 0.05. Cohen’s d was used to measure the effect size for treatments for power analysis purposes. Results: Statistically significant differences were found between the treatments OFF Med/ON DBS compared to ON Med/OFF DBS; and ON Med/ON DBS compared to ON Med/OFF DBS for the variables UPDRS, GDI, GPS and GVS (Gait Variable Score) (Hip Flexion/Extension, Knee Flexion/Extension). The effect size observed between treatments (ON Med/ON DBS versus ON Med/OFF DBS and ON Med/OFF DBS versus OFF Med/ON DBS) was high for both treatment comparisons for the variables UPDRS, GPS overall, GPS side and GVS (Hip Flexion/Extension and Knee Flexion/Extension). The effect size for GDI was medium for ON med/ON DBS versus ON Med/OFF DBS and high for the ON Med/OFF DBS versus OFF Med/ON DBS comparison (Cohen’s d = 0.45). Discussion and conclusions: There is evidence that the symptoms that are most responsive to dopamine are also those that provide improved results under the influence of stimulation. However, it is believed that a combination of DBS of the SNT and medication reduces motor fluctuations and dyskinesias, as well as reducing tremors, muscle stiffness and bradykinesia. The exact mechanisms of stimulation are still unknown. The effect of DBS on the STN during walking can be partially measured by the interaction of pedunculopontine nucleus (PPN) and STN. It is believed that the effects of treatment with levodopa were potentiated by the addition of stimulation, suggesting a synergistic effect of the two treatments by different routes. We observed that the best scores in UPDRS, GDI and GPS/MAP were obtained when patients were under the effect of two treatments together. Reference [1] Beynon S, McGinley L, Dobson F, Baker R. Correlations of the gait profile score and the movement analysis Profile relative to clinical judgments. Gait Posture 2010;32:129–32.
http://dx.doi.org/10.1016/j.gaitpost.2014.04.168
Fig. 1. GEI.
P57 An attempt to detect impairment by silhouette-based gait feature Chengju Zhou ∗ , Ikuhisa Mitsugami, Yasushi Yagi The Institute of Scientific and Industrial Research, Osaka University, 8-1 Mihogaoka, Ibaraki, Osaka 567-0047 Japan Introduction and aim: We can usually guess if a person has some impairments, such as a person whose leg is immobilized in a plastic cast, an elder person who cannot bend his/her knees, and a cataract patient, just by seeing them walking. If we can realize a system that can automatically estimate the impairment by observing them by a camera, it would be useful; e.g., in a commercial facility such a system would be helpful for staffs to find and assist a person with impairment immediately. There are several studies that investigate a difference of normal and abnormal walking, but they cannot be applied for our purpose since they use accurate 3-D pose obtained by expensive motion capture systems. Some other studies [1] use cameras but they use theatrical actions which are quite different from natural impaired walking. In our work, we apply a vision-based gait feature extraction technique that has been applied to personal authentication task for discrimination of normal and impaired walking. Patients/materials and methods: We need to collect data of walking of a huge number of subjects. It is, however, difficult to find so many subjects who really have impairments. In our study, therefore, we used knee supporters that restrict bending knees to simulate leg impairment, and a goggle glass which narrows their fields of view to act as visual impairment. Each subject walked a straight path, wearing the supporters on both legs, wearing the goggle, or wearing neither of them. The number of these three categories are 189, 142, and 235, respectively. A camera captured each subject from his/her side. We extracted his/her binary silhouette images from the captured movies by the background subtraction. We call this image sequence Gait Silhouette Volume (GSV). From the GSV, we calculated an averaged image which is called Gait Energy Image (GEI) [2] (Fig. 1) that is a feature representation originally proposed for human authentication. Principal Component Analysis (PCA) is then applied to the feature vector to reduce redundancy. To separate the normal walking and the impaired walking, Linear Discriminant Analysis (LDA) is used to find optimal discriminative axes. Once the axes are obtained, we re-project them to the original feature space. The re-projected features well describe dominant differences between the categories, which provide import hints for doctors to detect abnormal regions. Results: Fig. 2 shows experimental results. There are two comparisons; (a) normal and leg impaired, and (b) normal and visual impaired. For each case, classification accuracy is shown for quantitative evaluation. In addition, to illustrate the difference between normal and impaired walking, we show the re-projection of most discriminate vector calculated by LDA. Discussion and conclusions: From the results, we confirmed that it is possible to estimate existence of leg/visual impairment with promising accuracy using the GEI that is originally proposed
Abstracts / Gait & Posture 39S (2014) S1–S141
S123
Fig. 2. Classification results; confusion matrices and re-projection of most discriminate vector. (a) Normal and leg impairment; (b) Normal and visual impairment.
for gait authentication. Especially, visual impairment is easier to detect since most people with visual impairments bend their heads to the front to take care of their steps. Leg impairment is a little more difficult, but we find consistent difference in their leg motions. Future work contains increasing the number of subjects to achieve more reliable evaluation. Reference [1] Mostayed A. Abnormal gait detection using discrete Fourier transformation. In: Proc. international conference on multimedia and ubiquitous engineering. 2008. [2] Han J, Bhanu B. Individual recognition using gait energy image. IEEE Trans Pattern Anal Mach Intell 2006;28(February (2)):316–22.
http://dx.doi.org/10.1016/j.gaitpost.2014.04.169 P58
Table 1 Hip joint kinematics (degrees).
Extension Flexion Range (ext–flex) Adduktion Abduktion Range (add–abd) Inwards rotation Outward rotation Range (in–out rot)
Mean
∂
9.4 34.2 43.6 7.7 2.7 10.4 20.5 9.2 11.3
4.7 4.6 2.6 1.1 1.1 3 1.7 7.4 2.7
and order of measurement. The resolution of the measurements was defined as the 95% interval of the limits of the variation of the mean based on the formula (1). These limits correspond to the detection level of an established change with a probability of 95%.
Longitudinal test–retest of instrumented gait analysis using Lundberg skin marker model with focus on hip kinematic
√ ∂ = 1.96 · 2 ·
R. Zügner ∗ , J. Kärrholm, K. Pettersson, R. Tranberg
Results: In general the mean angular values of hip showed a variation corresponding to a detection level of less than 5◦ , except for outwards rotation where it amounted to 7.4◦ (Table 1). Discussion and conclusions: Pre- and postoperative clinical evaluation is frequently discussed and when it should be carried out. Together with the involving instrumental and methodological errors it can be difficult to predict changes. From the results from this study over a two-year period of 3DGA investigations with Lundberg skin marker model we can conclude that the over all variance (∂) is comparatively low.
Department of Orthopaedics, Institute of Clinical Sciences Sahlgrenska Academy, University of Gothenburg, Sahlgrenska University Hospital, SE 413 45 Göteborg, Sweden Introduction and aim: Three dimensional gait analyses (3DGA) are used to detected abnormalities or changes in gait and it are an important tool in research and for decision making in clinical treatment and evaluation. The rehabilitation time differs between treatment onset and when the evaluation with 3DGA should be carried out. To predict changes over time it is necessary to be aware of the size of instrumental and methodological errors when performing a gait analysis. The clinical outcome measurements from 3DGA are frequently discussed in investigation of patients with cerebral palsy. A number of factors may influence the outcome of a gait analysis. One of the major factors is the ability to repeat marker placements together with, how well the patient can reproduce their own gait pattern. Furthermore, the ability to track markers in the 3DGA-system is also of importance [1]. However, every laboratory is responsible to make test–retest measurement pointing out the involving error during the gait analyse procedure which can lead you to more secure interpretation [2,3]. The aim of our study was to evaluate the variation in the Lundberg skin marker model during a period of two years. Patients/materials and methods: Two healthy male subjects (mean age 41 years, height 1.82, 1.78 m, weight of 101 and 82 kg) volunteered in the study. 18 markers (∅19 mm) were attached with double-sided adhesive tape to the skin. Bony-landmarks were palpated at every session using an instruction guideline. A six-camera motion capture system (ProReflex MCU 240 Hz, Qualisys AB, Göteborg, Sweden) were used for data recording. During all recordings, subjects were asked to walk in self-selected speed. In total 400 gait trials, during a period of two years including left and right hip were examined. Analysis of variance and mixed models were used (SAS® ). Variables included in the model were person studied, side
2 + Skor
2 Sres 10
(1)
Reference [1] Wilken JM, et al. Reliability and minimal detectible change values for gait kinematics and kinetics in healthy adults. Gait Posture 2012;35(2):301–7. [2] Monaghan K, et al. Increasing the number of gait trial recordings maximises intra-rater reliability of the CODA motion analysis system. Gait Posture 2007;25(2):303–15. [3] Schwartz MH, et al. Measurement and management of errors in quantitative gait data. Gait Posture 2004;20(2):196–203.
http://dx.doi.org/10.1016/j.gaitpost.2014.04.170 P59 Visualisation of knee replacement rehabilitation exercises in the home Mobolaji Ayoade 1,∗ , Lynne Baillie 1 , Philip Rowe 2 , Tracey H. Howe 3 1 School of Engineering and Built Environment, Glasgow Caledonian University, UK 2 Bioengineering Department, Strathclyde University, UK 3 School of Health and Life Science, Glasgow Caledonian University, UK
Introduction and aim: Osteoarthritis of the knee is a degenerative joint disease associated with ageing, characterised by excessive
Abstracts / Gait & Posture 39S (2014) S1–S141
simple and easy to interpret measurements for clinical [1]. The aim of the present study was to analyze and compare the gait parameters of PD patients under the influence of dopaminergic medication and/or high frequency DBS using the GDI and the GPS/MAP. Patients/materials and methods: Sixteen patients with PD (eleven male and five female) who were submitted to bilateral high frequency DBS of the STN participated in the study. The gait assessment was conducted using three-dimensional kinematics (SMART-D® BTS) in three conditions: without medication and with stimulation (OFF med/ON DBS); with medication and stimulation (ON med/ON DBS); with medication and without stimulation (ON med/OFF DBS). The label of the markers and the processing of the biomechanical model to obtain kinematic data were performed using Vicon Nexus® software and the Plug in Gait® model. The kinematic data were imported into a spreadsheet, where a mathematical routine was used to calculate the GDI and GPS/MAP in different conditions. The Unified Parkinson’s Disease Rating Scale (UPDRS) part III also was applied during the three conditions. The data were analyzed using the variance for repeated measures test (ANOVA), with the level of statistical significance set at p < 0.05. Cohen’s d was used to measure the effect size for treatments for power analysis purposes. Results: Statistically significant differences were found between the treatments OFF Med/ON DBS compared to ON Med/OFF DBS; and ON Med/ON DBS compared to ON Med/OFF DBS for the variables UPDRS, GDI, GPS and GVS (Gait Variable Score) (Hip Flexion/Extension, Knee Flexion/Extension). The effect size observed between treatments (ON Med/ON DBS versus ON Med/OFF DBS and ON Med/OFF DBS versus OFF Med/ON DBS) was high for both treatment comparisons for the variables UPDRS, GPS overall, GPS side and GVS (Hip Flexion/Extension and Knee Flexion/Extension). The effect size for GDI was medium for ON med/ON DBS versus ON Med/OFF DBS and high for the ON Med/OFF DBS versus OFF Med/ON DBS comparison (Cohen’s d = 0.45). Discussion and conclusions: There is evidence that the symptoms that are most responsive to dopamine are also those that provide improved results under the influence of stimulation. However, it is believed that a combination of DBS of the SNT and medication reduces motor fluctuations and dyskinesias, as well as reducing tremors, muscle stiffness and bradykinesia. The exact mechanisms of stimulation are still unknown. The effect of DBS on the STN during walking can be partially measured by the interaction of pedunculopontine nucleus (PPN) and STN. It is believed that the effects of treatment with levodopa were potentiated by the addition of stimulation, suggesting a synergistic effect of the two treatments by different routes. We observed that the best scores in UPDRS, GDI and GPS/MAP were obtained when patients were under the effect of two treatments together. Reference [1] Beynon S, McGinley L, Dobson F, Baker R. Correlations of the gait profile score and the movement analysis Profile relative to clinical judgments. Gait Posture 2010;32:129–32.
http://dx.doi.org/10.1016/j.gaitpost.2014.04.168
Fig. 1. GEI.
P57 An attempt to detect impairment by silhouette-based gait feature Chengju Zhou ∗ , Ikuhisa Mitsugami, Yasushi Yagi The Institute of Scientific and Industrial Research, Osaka University, 8-1 Mihogaoka, Ibaraki, Osaka 567-0047 Japan Introduction and aim: We can usually guess if a person has some impairments, such as a person whose leg is immobilized in a plastic cast, an elder person who cannot bend his/her knees, and a cataract patient, just by seeing them walking. If we can realize a system that can automatically estimate the impairment by observing them by a camera, it would be useful; e.g., in a commercial facility such a system would be helpful for staffs to find and assist a person with impairment immediately. There are several studies that investigate a difference of normal and abnormal walking, but they cannot be applied for our purpose since they use accurate 3-D pose obtained by expensive motion capture systems. Some other studies [1] use cameras but they use theatrical actions which are quite different from natural impaired walking. In our work, we apply a vision-based gait feature extraction technique that has been applied to personal authentication task for discrimination of normal and impaired walking. Patients/materials and methods: We need to collect data of walking of a huge number of subjects. It is, however, difficult to find so many subjects who really have impairments. In our study, therefore, we used knee supporters that restrict bending knees to simulate leg impairment, and a goggle glass which narrows their fields of view to act as visual impairment. Each subject walked a straight path, wearing the supporters on both legs, wearing the goggle, or wearing neither of them. The number of these three categories are 189, 142, and 235, respectively. A camera captured each subject from his/her side. We extracted his/her binary silhouette images from the captured movies by the background subtraction. We call this image sequence Gait Silhouette Volume (GSV). From the GSV, we calculated an averaged image which is called Gait Energy Image (GEI) [2] (Fig. 1) that is a feature representation originally proposed for human authentication. Principal Component Analysis (PCA) is then applied to the feature vector to reduce redundancy. To separate the normal walking and the impaired walking, Linear Discriminant Analysis (LDA) is used to find optimal discriminative axes. Once the axes are obtained, we re-project them to the original feature space. The re-projected features well describe dominant differences between the categories, which provide import hints for doctors to detect abnormal regions. Results: Fig. 2 shows experimental results. There are two comparisons; (a) normal and leg impaired, and (b) normal and visual impaired. For each case, classification accuracy is shown for quantitative evaluation. In addition, to illustrate the difference between normal and impaired walking, we show the re-projection of most discriminate vector calculated by LDA. Discussion and conclusions: From the results, we confirmed that it is possible to estimate existence of leg/visual impairment with promising accuracy using the GEI that is originally proposed
Abstracts / Gait & Posture 39S (2014) S1–S141
S123
Fig. 2. Classification results; confusion matrices and re-projection of most discriminate vector. (a) Normal and leg impairment; (b) Normal and visual impairment.
for gait authentication. Especially, visual impairment is easier to detect since most people with visual impairments bend their heads to the front to take care of their steps. Leg impairment is a little more difficult, but we find consistent difference in their leg motions. Future work contains increasing the number of subjects to achieve more reliable evaluation. Reference [1] Mostayed A. Abnormal gait detection using discrete Fourier transformation. In: Proc. international conference on multimedia and ubiquitous engineering. 2008. [2] Han J, Bhanu B. Individual recognition using gait energy image. IEEE Trans Pattern Anal Mach Intell 2006;28(February (2)):316–22.
http://dx.doi.org/10.1016/j.gaitpost.2014.04.169 P58
Table 1 Hip joint kinematics (degrees).
Extension Flexion Range (ext–flex) Adduktion Abduktion Range (add–abd) Inwards rotation Outward rotation Range (in–out rot)
Mean
∂
9.4 34.2 43.6 7.7 2.7 10.4 20.5 9.2 11.3
4.7 4.6 2.6 1.1 1.1 3 1.7 7.4 2.7
and order of measurement. The resolution of the measurements was defined as the 95% interval of the limits of the variation of the mean based on the formula (1). These limits correspond to the detection level of an established change with a probability of 95%.
Longitudinal test–retest of instrumented gait analysis using Lundberg skin marker model with focus on hip kinematic
√ ∂ = 1.96 · 2 ·
R. Zügner ∗ , J. Kärrholm, K. Pettersson, R. Tranberg
Results: In general the mean angular values of hip showed a variation corresponding to a detection level of less than 5◦ , except for outwards rotation where it amounted to 7.4◦ (Table 1). Discussion and conclusions: Pre- and postoperative clinical evaluation is frequently discussed and when it should be carried out. Together with the involving instrumental and methodological errors it can be difficult to predict changes. From the results from this study over a two-year period of 3DGA investigations with Lundberg skin marker model we can conclude that the over all variance (∂) is comparatively low.
Department of Orthopaedics, Institute of Clinical Sciences Sahlgrenska Academy, University of Gothenburg, Sahlgrenska University Hospital, SE 413 45 Göteborg, Sweden Introduction and aim: Three dimensional gait analyses (3DGA) are used to detected abnormalities or changes in gait and it are an important tool in research and for decision making in clinical treatment and evaluation. The rehabilitation time differs between treatment onset and when the evaluation with 3DGA should be carried out. To predict changes over time it is necessary to be aware of the size of instrumental and methodological errors when performing a gait analysis. The clinical outcome measurements from 3DGA are frequently discussed in investigation of patients with cerebral palsy. A number of factors may influence the outcome of a gait analysis. One of the major factors is the ability to repeat marker placements together with, how well the patient can reproduce their own gait pattern. Furthermore, the ability to track markers in the 3DGA-system is also of importance [1]. However, every laboratory is responsible to make test–retest measurement pointing out the involving error during the gait analyse procedure which can lead you to more secure interpretation [2,3]. The aim of our study was to evaluate the variation in the Lundberg skin marker model during a period of two years. Patients/materials and methods: Two healthy male subjects (mean age 41 years, height 1.82, 1.78 m, weight of 101 and 82 kg) volunteered in the study. 18 markers (∅19 mm) were attached with double-sided adhesive tape to the skin. Bony-landmarks were palpated at every session using an instruction guideline. A six-camera motion capture system (ProReflex MCU 240 Hz, Qualisys AB, Göteborg, Sweden) were used for data recording. During all recordings, subjects were asked to walk in self-selected speed. In total 400 gait trials, during a period of two years including left and right hip were examined. Analysis of variance and mixed models were used (SAS® ). Variables included in the model were person studied, side
2 + Skor
2 Sres 10
(1)
Reference [1] Wilken JM, et al. Reliability and minimal detectible change values for gait kinematics and kinetics in healthy adults. Gait Posture 2012;35(2):301–7. [2] Monaghan K, et al. Increasing the number of gait trial recordings maximises intra-rater reliability of the CODA motion analysis system. Gait Posture 2007;25(2):303–15. [3] Schwartz MH, et al. Measurement and management of errors in quantitative gait data. Gait Posture 2004;20(2):196–203.
http://dx.doi.org/10.1016/j.gaitpost.2014.04.170 P59 Visualisation of knee replacement rehabilitation exercises in the home Mobolaji Ayoade 1,∗ , Lynne Baillie 1 , Philip Rowe 2 , Tracey H. Howe 3 1 School of Engineering and Built Environment, Glasgow Caledonian University, UK 2 Bioengineering Department, Strathclyde University, UK 3 School of Health and Life Science, Glasgow Caledonian University, UK
Introduction and aim: Osteoarthritis of the knee is a degenerative joint disease associated with ageing, characterised by excessive