UserMicrosoft Word - JAEBS-1389-9, zairiJ. Appl. Environ. Biol. Sci., 7(1)150-156, 2017 |
© 2017, TextRoad Publication |
ISSN: 2090-4274 |
Journal of Applied Environmental and Biological Sciences www.textroad.com |
Automated Classification of Autism Spectrum Disorders Gait Patterns |
Using Discriminant Analysis Based on Kinematic and Kinetic Gait Features |
Che Zawiyah Che Hasan1, Rozita Jailani1, Nooritawati Md Tahir1, Ihsan Mohd Yassin1, Zairi Ismael Rizman2 |
1Faculty of Electrical Engineering, Universiti Teknologi MARA, Shah Alam, Selangor, Malaysia |
2Faculty of Electrical Engineering, Universiti Teknologi MARA, Dungun, Terengganu, Malaysia |
Received: July 3, 2016 |
Accepted: October 27, 2016 |
ABSTRACT |
Autism spectrum disorder (ASD) is a permanent neurodevelopmental disorder that can be recognised during the |
first few years of life and is further supported by the existence of gait impairments. Automated classification of ASD gait could provide assistance in diagnosis and ensure rapid quantitative clinical judgement. This study proposes an automated classification of ASD gait patterns based on kinematic and kinetic gait features with the application of machine learning approaches. Gait analysis of 24 ASD and 24 typical healthy children were recorded using a state-of-the-art three-dimensional (3D) motion analysis system and two force platforms during barefoot self-selected normal walking. Nine kinematic and sixteen kinetic gait features were statistically selected using the independent t-tests and Mann-Whitney U tests, which grouped into two types of datasets. Linear discriminant analysis (LDA) and quadratic discriminant analysis (QDA) were employed to perform the recognition task. Overall, the results of the proposed study suggest that LDA classifier with kinetic gait features as input predictors produces better classification performance with 82.50% of accuracy and lower misclassification rate. |
KEYWORDS: Autism Spectrum Disorder (ASD), Linear Discriminant Analysis (LDA), Quadratic |
Discriminant Analysis (QDA), Gait Analysis, Gait Classification, Gait Features. |
INTRODUCTION |
Autism spectrum disorder (ASD) is a condition that can be characterised by a constant deficit in social |
communication, social interaction and the presence of restrictive and repetitive behaviour. It is a lifelong neurodevelopmental disorder that can be recognised during the early stage of the developmental period of a child. Other additional characteristics that support the diagnosis of ASD are the existence of motor deficits which include abnormal gait, clumsiness and irregular motor signs [1]. Abnormal gait is defined as an unusual style of walking from the normal walking pattern and this could cause deterioration in occupational and other substantial ranges of daily activities. |
A number of studies have addressed various types of gait disturbance in children with ASD [2]-[6]. These studies had successfully documented abnormal gait features in individuals with ASD which could be beneficial for early diagnosis and better treatment planning for children with ASD [7]. Conventional methods of gait assessment are often time-consuming and highly dependent on the clinician judgement, which leads to subjective and qualitative evaluations. With the current advances in gait analysis technology, which not only allow new insights in understanding normal and pathological movement patterns but also provide prospects to develop an automated diagnosis of pathological disorders. |
The use of machine learning classifiers for automated recognition of gait pattern deviations has grown enormously in the last decades. Artificial neural network (ANN) and support vector machines (SVM) have been employed for recognition and classification of Parkinson’s disease [8], young-old gait patterns [9], cerebral palsy children [10] and patients with neurological disorders [11]. ANN was also successfully used for classification of gender in children [12] and post-stroke patients [13] based on their gait profile. More recent studies have applied linear discriminant analysis (LDA) to extract dominant features for discrimination of Parkinson’s disease [14]. Most of the mentioned studies have used either basic gait measurements, kinematic, kinetic or combination of those features to be fed into machine classifiers. |
Until now, the published literature dealing with automated classification on ASD gait patterns is, however scarce. Thus, this study proposes an automated classification of gait patterns of children with ASD based on kinematic and kinetic gait features. These features were first extracted using waveform analyses and parameterisation methods and then were selected using statistical techniques. Linear discriminant analysis (LDA) and quadratic discriminant analysis (QDA) were employed to model both input features and their classification performances with each input dataset were compared. |
Corresponding Author: Rozita Jailani, Faculty of Electrical Engineering, Universiti Teknologi MARA, Shah Alam, |
Selangor, Malaysia, E-mail: rozita@ieee.org |
This paper is organised as follows. The methodology of the proposed automatic gait classification will be |
elaborated in Methodology section. The obtained results will be presented and further discussed in Results and Discussions section. Finally, the summary of this study is presented in the Conclusion section. |
METHODOLOGY |
The methodology for the proposed gait classification system is illustrated in a block diagram as shown in |
Figure 1. It consists of four stages which include gait data acquisition, pre-processing and feature extraction, feature selection statistical procedures, and ends with gait classification using two types of discriminant analysis. Finally, the performance of each model using two types of data sets was evaluated. |
Figure 1: Block diagram of the proposed methodology |
Gait Data Acquisition |
Two groups of 24 children with ASD (18 male, 6 female) and 24 healthy children (12 male, 12 female) participated in the study. Participants were in the range of 4 to 12 years old. Table 1 shows a summary of the participants’ characteristics. |
Table 1: Characteristics of group participants |
Characteristics | ASD | Control |
Number of participants | 24 | 24 |
Gender (Male:Female) | 18:6 | 12:12 |
Age (years) | 8.95(2.13) | 9.67(1.88) |
Height (m) | 1.31(0.14) | 1.27(0.13) |
Body mass (kg) | 32.71(15.24) | 27.83(8.95) |
Body Mass Index (BMI) | 18.41(5.63) | 16.78(2.66) |
Note: Data are given as total number and mean (standard deviation). |
The children with ASD were recruited from the National Autism Society of Malaysia (NASOM) centre in |
Klang, Selangor, Malaysia and local community by approaching the parents via social media network. All ASD participants had been diagnosed with a mild category by qualified medical practitioners and were able to follow verbal instructions. The ASD participants were attended either by their parents or caretakers during the experiment. The typical healthy children were recruited from the families of faculty members and neighbourhoods nearby. All participants were able to walk independently without any assistive devices and had no medical history of lower extremity injuries. An informed consent form was obtained from the parent or guardian of the participants. The research procedures have been approved by the Research Ethics Committee of Universiti Teknologi MARA (UiTM), Shah Alam, Selangor. |
The acquisition of gait data was performed at the Human Motion Gait Analysis laboratory at the UiTM Shah Alam using a state-of-the-art three dimensional (3D) motion analysis system. The system is equipped with an eight-camera (Vicon T-series) motion capture (Vicon Motion Systems Ltd., Oxford, United Kingdom) and two force plates (Advanced Mechanical Technology Inc., MA, USA). The 3D motion analysis system was used to record the movement of marker trajectories at 100 Hz and the force plates which were embedded in the middle of a 6.5-metre walkway were used to measure the 3D ground reaction forces (GRF) at 1000 Hz. A digital weight scale, measuring tape and a calliper was used to obtain anthropometric measurements of the participants. |
Thirty-five spherical retroreflective markers bilaterally adhered onto the specific anatomical landmarks of the participants by the same investigator, using the full body Plug-in Gait model (Vicon) based on the Newington-Helen Hayes gait model [15-16]. After static calibration, participants were instructed to perform a straight barefoot walking along the walkway at their self-selected speed. Participants were allowed to perform several practice trials for familiarisation and then an average of five to ten walking trials were recorded from each participant. However, several ASD participants with small stride length and sensory issues had to perform more than twenty trials in order to achieve a valid walking trial with a single foot contact on each force plate as in Figure 2. Both force plates were calibrated at the beginning of each trial to reduce the effect of noise from surroundings. |
Figure 2: A valid walking trial with a single foot contact on each force plate |
Data Pre-Processing and Features Extraction |
Pre-processing of walking trials involved the process of reviewing the entire trials obtained from each participant. However, due to several conditions such as sensory issues and small stride length of several ASD participants, only a single valid trial with a complete temporal-spatial, kinematic and kinetic measurements from each participant was chosen for further processing. |
The marker trajectories were smoothed using the built-in Woltring generalised cross-validatory spline algorithm to minimise trajectory noise [17]. The 3D ground reaction forces (GRF) from the force plate were filtered using a second-order low-pass Butterworth filter with 30 Hz cut-off frequency and then down-sampled to match the kinematics of 3D joint angles data [18]. The 3D GRFs, together with the anthropometric measurements and accelerations data, then inputted into standard inverse dynamics approach to providing the resultant 3D joint moments and joint powers for the hip, knee, and ankle joints [19]. All data pre-processing were computed using the Vicon Nexus software (version 1.8.5, Vicon, Oxford, UK). |
After pre-processing of the selected valid trials, then the kinematic and kinetic gait data from a single left limb gait cycle of the trial were chosen to represent each participant [20-21]. Furthermore, the children with ASD were found to demonstrate a typical interlimb movement during gait [22]. |
The gait cycle for every participant was normalised to the percentage of their stride duration. While the time component of the 3D GRFs was normalised to the percentage of stance phase time and the magnitude of GRF component was standardised to the percentage of participant’s body weight. The joint moments was normalised |
to body weight. These normalisation steps were carried out in order to eliminate variations among the |
participants with different height, body mass and duration of the gait cycle and stance phase time [23-24]. |
In this study, 12-kinematic waveforms were calculated in the sagittal, frontal and transverse plane for pelvis, hip, knee, ankle and 12-kinetic waveforms were examined in 3D joint moments for hip, knee, ankle and joint powers for hip, knee and ankle. To group the data from the 48 participants, each waveform was normalised to 51-time points, one for every 2% of the gait cycle [25-26]. Then, the maximum and minimum value from all waveforms, and sagittal joint angle at hip, knee, and ankle during foot contact and toe off were extracted [5]. For the 3D GRFs in medial-lateral (x-axis), anterior-posterior (y-axis), and vertical (z-axis) directions, parameterization techniques were applied to each waveform to extract the instantaneous values of amplitude and its corresponding time of occurrence [27]-[29]. Both normalisation and gait feature extraction were computed in Microsoft Excel version 2013 (Microsoft Corp., USA). |
Features Selection |
Practically, all extracted gait features could be used as input data in the classification process. However, there were conditions that some features were irrelevant or unrelated to be used in the process and it sometimes could deteriorate the generalised performances of the classifiers. One of the suggested solutions to overcome this issue is by performing features selection so that only dominant features could be used to represent each group. |
Features selection is also an important step in classification system in order to achieve good classification results especially if it deals with high dimensionality of data. By conducting this feature selection, the original high dimensional gait features could be reduced and only the significant features were selected as a dataset for classification stage. The gait features that were irrelevant were removed. |
In this stage, a between-group test was employed to all extracted features to select the significant gait features that would possibly discriminate the gait patterns between both groups. For this purpose, two types of parametric and non-parametric statistical procedures were implemented. These tests were chosen for the ability to provide significant between-group differences for both kinematic and kinetic gait features [30]. Furthermore, the preliminary study had shown that there was no significant difference for basic temporal and spatial gait parameters between both groups. |
Prior to analysis, each extracted gait features were explored for normality using the Shapiro-Wilk (SW) test [31]. The independent t-tests were used to compare the mean differences of gait features when data were normally distributed or parametric, and Mann-Whitney U tests for data that were non-parametric. The significant difference between the two groups was defined as p < 0.05. |
Classification Model |
Generally, classification is a machine learning technique used to categorise items in a collection to target classes or groups. In this study, the dominant kinematic and kinetic gait features that were selected from the feature selection stage were grouped into two types of data sets. These data sets were then used as input data to the machine classifiers. For this purpose, two types of discriminant analysis (DA), LDA and QDA were employed for classification purposes. DA is known as a simple classifier that is said to have low variance, mathematically robust, produces low error rates and accuracy as good as other complex classifiers [32-33]. |
DA algorithm assumes that different classes generate data based on different Gaussian distributions. To train a classifier, the fitting function evaluates the Gaussian distribution parameters for each class and then the created classifier finds the class with the smallest misclassification cost for predicting the classes of new data. For LDA, the model has the same covariance matrix but different means for each class. Meanwhile, for QDA, each class has different means and covariance matrixes. These two types of DA differ in the resultant discriminant function that is linear or quadratic. Further explanations on both classifiers were elucidated in [34-35]. |
This study employs a 10-fold cross validation method to test the generalisation ability of the classification method using various combinations of the testing and training data sets. These standard procedure has been used in a number of classification evaluations, particularly for data sets with small sample size [36]-[40]. |
In this method, the data set is randomly partitioned into 10 folds. A different fold of the data is used for testing, while the remaining nine folds are used as training data. 10-fold cross validation was iterated for ten times and the results from the folds are then averaged to produce the final result [41]. |
The performance of LDA and QDA classifiers with different input data of kinematic and kinetic data sets were measured using the accuracy and error rate based on the confusion matrix with two classes, ASD and normal. The accuracy is defined as the ratio of correctly classified cases to a total number of cases and error rate is the misclassification rate as given by Equation (1) and (2). |
Accuracy = (TP+TN)/(Total data) x 100% (1) |
Error rate = (FP+FN)/(Total data) (2) |
In this study, true positive (TP) is the number of ASD cases correctly classified and true negative (TN) is |
the number of normal cases correctly classified. False positive (FP) is the number of normal cases incorrectly classified as ASD and false negative (FN) is the number of ASD cases incorrectly classified as normal. |
RESULTS AND DISCUSSION |
The developed features extraction, features selection, and the effectiveness of the proposed ASD gait |
patterns classification using LDA and QDA are elaborated in this section. Based on waveform analyses and parameterization techniques employed in features extraction stage, a total of 34 kinematic features and 44 kinetic features were extracted to represent the gait profiles of all participants. |
Table 2 presents the kinematic and kinetic gait features selected using statistical feature selection techniques. The features with a p-value of less than 0.05 are assumed to provide great influence during the classification process. Based on the tabulated results, 9 joint angles at pelvis, hip, knee and ankle were found to be significant and these features represented kinematic dataset. Meanwhile, 5 joint moments, 4 joint powers and 7 from GRFs were clustered together as a kinetic dataset of 16 gait features. |
Table 3 summarises the classification performance in terms of the accuracy and the error rate for LDA and QDA classifiers using kinematic and kinetic data sets selected using statistical feature selection techniques. Based on the two types of datasets as input features for both classifiers, the combination of the kinetic dataset and LDA classifier demonstrates a good performance for ASD gait classification with 82.50% of accuracy and lower misclassification rate of 0.175. Results also indicate that kinetic dataset is more discriminatory than the kinematic dataset. In other words, the ASD and normal groups differ mainly in the kinetic gait features. These results are expected since most of the gait deviations observed in ASD group is mostly signified by the kinetic gait features. Future studies should implement another type of feature selection techniques and other machine classifiers to optimise the classification accuracy. |
Table 2: Kinematic and kinetic gait features selected using statistical feature selection techniques |
Kinematic Gait Features (9) | | Kinetic Gait Features (16) |
3D Joint Angles | 3D Joint Moments | Joint Powers | 3D GRFs |
Max pelvic tilt
Min pelvic obliquity
Hip flex foot contact
Max hip extension (stance)
Hip extension foot off
Knee flexion foot contact
Max knee abduction
Max ankle dorsiflexion (stance)
Max ankle plantarflexion (swing) | Max knee adduction moment Y
Max knee rotation moment Z
Max ankle plantar flexor moment X
Max ankle moment Z
Min ankle moment Z | Max hip power
Min hip power
Min knee power
Max eccentric ankle power | Fx3
Fy1
Fy3
Fz3
Ty2
Push-off rate
Peak ratio |
Table 3: Classification accuracy and error rate for LDA and QDA classifiers using kinematic and |
kinetic data sets |
Dataset | Classifier | Accuracy (%) | Error Rate |
Kinematic | LDA | 70.00 | 0.300 |
QDA | 80.00 | 0.200 |
Kinetic | LDA | 82.50 | 0.175 |
| QDA | 71.67 | 0.283 |
CONCLUSION |
In this paper, discriminant analysis classifier for classifying gait patterns of children with ASD based on |
kinematic and kinetic gait features has been proposed. This study introduces statistical techniques for selecting dominant gait features from gait data obtained from a state-of-the-art 3D motion analysis system. In recognising gait impairments in children with ASD, two types of gait datasets had been discovered. Overall the results of this study suggest that LDA classifier with kinetic gait features as input predictors is more effective for categorising gait patterns of children with ASD. These potential findings would be beneficial for future applications in identifying gait abnormalities in individuals with ASD or other pathological gait patterns. |
ACKNOWLEDGMENT |
The authors wish to thank the Ministry of Higher Education (MOHE), Malaysia for the funds received |
through the Niche Research Grant Scheme (NRGS), project file: 600-RMI/NRGS 5/3 (8/2013), the Human Motion Gait Analysis Laboratory, Institute of Research Management and Innovation Unit (IRMI) Premier Laboratory, Universiti Teknologi MARA (UiTM) Shah Alam, Selangor. The authors also wish to acknowledge |
all contributions especially the National Autism Society of Malaysia (NASOM), volunteered participants and |
their families. This study was undertaken during study leave of the main author under the 2014 Federal Training (HLP) sponsorship scheme awarded by the MOHE, Malaysia. |
REFERENCES |
1. American Psychiatric Association, 2013. Diagnostic and statistical manual of mental disorders. American |
Psychiatric Association. |
2. Teitelbaum, P., O. Teitelbaum, J. Nye, J. Fryman and R.G. Maurer, 1998. Movement Analysis in Infancy May Be Useful for Early Diagnosis of Autism. Proceedings of the National Academy of Sciences, 95 (23): 13982-13987. |
3. Vilensky, J.A., A.R. Damasio and R.G. Maurer, 1981. Gait Disturbances in Patients With Autistic Behavior: A Preliminary Study. Archives of Neurology, 38 (1): 646-649. |
4. Ambrosini, D., E. Courchesne and K. Kaufman, 1998. Motion Analysis of Patients With Infantile Autism. Gait Posture, 7 (2): 188. |
5. Calhoun, M., M. Longworth and V.L. Chester, 2011. Gait Patterns in Children With Autism. Clinical Biomechanics, 26 (2): 200-206. |
6. Weiss, M.J., M.F. Moran, M.E. Parker and J.T. Foley, 2013. Gait Analysis of Teenagers and Young Adults Diagnosed With Autism and Severe Verbal Communication Disorders. Frontiers in Integrative Neuroscience, 7 (33): 1-10. |
7. Kindregan, D., L. Gallagher and J. Gormley, 2015. Gait Deviations in Children With Autism Spectrum Disorders: A Review. Autism Research and Treatment, 2015: 1-8. |
8. Tahir, N.M. and H.H. Manap, 2012. Parkinson Disease Gait Classification Based on Machine Learning Approach. Journal of Applied Science, 12 (2): 180-185. |
9. Begg, R.K., M. Palaniswami and B. Owen, 2005. Support Vector Machines for Automated Gait Classification. IEEE Transactions on Biomedical Engineering, 52 (5): 828-838. |
10. Kamruzzaman, J. and R.K. Begg, 2006. Support Vector Machines and Other Pattern Recognition Approaches to the Diagnosis of Cerebral Palsy Gait. IEEE Transactions on Biomedical Engineering, 53 (12): 2479-2490. |
11. Pradhan, C., M. Wuehr, F. Akrami, M. Neuhaeusser, S. Huth, T. Brandt, K. Jahn and R. Schniepp, 2015. Automated Classification of Neurological Disorders of Gait Using Spatio-Temporal Gait Parameters. Journal of Electromyography and Kinesiology, 25 (2): 413-422. |
12. Zakaria, N.K., R. Jailani and N.M. Tahir, 2015. Application of ANN in Gait Features of Children for Gender Classification. Procedia Computer Science, 76: 235-242. |
13. Kaczmarczyk, K., A. Wit, M. Krawczyk and J. Zaborski, 2009. Gait Classification in Post-Stroke Patients Using Artificial Neural Networks. Gait Posture, 30 (2): 207-210. |
14. Ilias, S., R. Jailani and N.M. Tahir, 2015. Comparison Between Linear Discriminant Analysis and Singular Value Decomposition for PD Gait Classification. In the Proceedings of the 2015 IEEE Symposium on Computer Applications and Industrial Electronics, pp: 142-146. |
15. Kadaba, M.P., H.K. Ramakrishnan and M E. Wootten, 1990. Measurement of Lower Extremity Kinematics During Level Walking. Journal of Orthopaedic Research, 8 (3): 383-392. |
16. Davis, R.B., S. Ounpuu, D. Tyburski and J.R. Gage, 1991. A Gait Analysis Data Collection and Reduction Technique. Human Movement Science, 10 (5): 575-587. |
17. Woltring, H.J., 1986. A Fortran Package for Generalized, Cross-Validatory Spline Smoothing and Differentiation. Advances in Engineering Software, 8 (2): 104-113. |
18. Thompson, M.A., S.S. Lee, J. Seegmiller and C.P. McGowan, 2015. Kinematic and Kinetic Comparison of Barefoot and Shod Running in Mid/Forefoot and Rearfoot Strike Runners. Gait Posture, 41 (4): 957-959. |
19. Christopher L. Vaughan, Brian L. Davis and Jeremy C. O’Connor, 1999. Dynamics of human gait. Kiboho Publishers. |
20. Massaad, A., A. Assi, W. Skalli and I. Ghanem, 2014. Repeatability and Validation of Gait Deviation Index |
in Children: Typically Developing and Cerebral Palsy. Gait Posture, 39 (1): 354-358. |
21. Dixon, P.C., M.V. Bowtell and J. Stebbins, 2014. The Use of Regression and Normalisation for the Comparison of Spatio-Temporal Gait Data in Children. Gait Posture, 40 (4): 521-525. |
22. Chester, V.L. and M. Calhoun, 2012. Gait Symmetry in Children With Autism. Autism Research and Treatment, 2012: 1-5. |
23. Michael W. Whittle, 2014. Gait analysis: An introduction. Butterworth-Heinemann. |
24. Stansfield, B.W., S.J. Hillman, M.E. Hazlewood, A.M. Lawson, A.M. Mann, I.R. Loudon and J.E. Robb, 2003. Normalisation of Gait Data in Children. Gait Posture, 17 (1): 81-87. |
25. Tingley, M., C. Wilson, E. Biden and W.R. Knight, 2002. An Index to Quantify Normality of Gait in Young Children. Gait Posture, 16 (2): 149-158. |
26. Chester, V.L. and A.T. Wrigley, 2008. The Identification of Age-Related Differences in Kinetic Gait Parameters Using Principal Component Analysis. Clinical Biomechanics, 23 (2): 212-220. |
27. Agouris, I., 2002. The variability of force platform, kinematic and kinetic data in normal and cerebral palsy gait, Phd thesis, University of Aberdeen, Scotland. |
28. Su, B.L., R. Song, L.Y. Guo and C.W. Yen, 2015. Characterizing Gait Asymmetry Via Frequency Sub- Band Components of the Ground Reaction Force. Biomedical Signal Processing and Control, 18: 56-60. |
29. McCrory, J.L., A.J. Chambers, A. Daftary and M.S. Redfern, 2011. Ground Reaction Forces During Gait in Pregnant Fallers and Non-Fallers. Gait Posture, 34 (4): 524-528. |
30. Allison, K., T.V. Wrigley, B. Vicenzino, K.L. Bennell, A. Grimaldi and P.W. Hodges, 2016. Kinematics and Kinetics During Walking in Individuals With Gluteal Tendinopathy. Clinical Biomechanics, 32: 56-63. |
31. A. Mayers, 2013. Introduction to statistics and SPSS in psychology. Pearson Education. |
32. Sankar K. Pal and P. Mitra, 2004. Pattern Recognition algorithms for data mining: Scalability, knowledge discovery and soft granular computing. CRC Press. |
33. Pang N. Tan, 2006. Introduction to data mining. Pearson Education India. |
34. Kim, K.S., H.H. Choi, C.S. Moon and C.W. Mun, 2011. Comparison of K-Nearest Neighbor, Quadratic Discriminant and Linear Discriminant Analysis in Classification of Electromyogram Signals Based on the Wrist-Motion Directions. Current Applied Physics, 11 (3): 740-745. |
35. Omar, T., Z. Wassim and B.M. Mohamed, 2014. Brain-Computer Interface: Frequency Domain Approach Using the Linear and the Quadratic Discriminant Analysis. In the Proceedings of the 2014 1st International Conference on Advanced Technologies for Signal and Image Processing, pp: 346-349. |
36. Begg, R. and J. Kamruzzaman, 2005. A Machine Learning Approach for Automated Recognition of Movement Patterns Using Basic, Kinetic and Kinematic Gait Data. Journal of Biomechanics, 38 (3): 401- 408. |
37. Manap, H.H., N.M. Tahir and A.I.M. Yassin, 2011. Statistical Analysis of Parkinson Disease Gait Classification Using Artificial Neural Network. In the Proceedings of the 2011 IEEE International Symposium on Signal Processing and Information Technology, pp: 60-65. |
38. Er, F.Ç., Z. Firat, I. Kovanlikaya, U. Ture and E. Ozturk-Isik, 2013. Multivariate Discriminant Analysis of Multiparametric Brain MRI to Differentiate High Grade and Low Grade Gliomas-A Computer-Aided Diagnosis Development Study. In the Proceedings of the 2013 IEEE 13th International Conference on Bioinformatics and Bioengineering, pp: 1-5. |
39. Zhang, L., O. Ma, J.M. Fabre, R.H. Wood, S.U. Garcia, K.M. Ivey and E D. Mccann, 2015. Classification of Older Adults With/Without a Fall History Using Machine Learning Methods. In the Proceedings of the 2015 37th Annual International Conference of the IEEE Engineering in Medicine and Biology Society, pp: 6760-6763. |
40. Wahid, F., R.K. Begg, C.J. Hass, S. Halgamuge and D.C. Ackland, 2015. Classification of Parkinson’s Disease Gait Using Spatial-Temporal Gait Features. IEEE Journal of Biomedical and Health Informatics, 19 (6): 1794-1802. |
41. Kohavi, R., 1995. A Study of Cross-Validation and Bootstrap for Accuracy Estimation and Model Selection. International Joint Conference on Artificial Intelligence, 14 (2): 1137-1145. |