1. Mobility-Based Models for Advancing Diagnostic/Predictive Healthcare
Student Research and Activity Fair 2019
Mobility-Based Models for Advancing Diagnostic/Predictive Healthcare
Elham Rastegari
Adviser:
Hesham Ali
You have 35 ‘real’ slides (not counting navigation and title slide). This is too much for 35 mins. Try to condense to ‘real’ slides.

2. Introduction Movement, gait and health conditions
Research Foundations
Research Method
Completed Studies
Conclusion
Introduction
Movement, gait and health conditions
Traditional movement measurement systems
Advanced technologies
Continuously and remotely monitoring
Disadvantages of traditional movement measurement systems: expensive, requires individuals to come to laboratory, limited number of trials, people may perform differently when they are supervised, does not allow continuous monitoring of individuals
Advantages of traditional movement measurement systems: accurate, and already validated results
Advantages of wearable devices: cheap, allow continuous monitoring in the free-living condition
Disadvantages of wearable devices: not as accurate as traditional movement measurement systems
11/18/2019

3. Introduction
Research Foundations
Research Method
Completed Studies
Conclusion
Introduction
long-standing reactive treatment approach to the early detection and preventative era.
Wearable-based movement data, when combined with other relevant data could enhance evidence-based medicine.
Not utilized in real-life applications.
Q: Utilizing movement parameters and new diagnostic/ prognostic models, how can we promote decision-making process in the healthcare domain?
There are huge number of paper on gait and health conditions using traditional movement analysis techniques such as statistical analysis (ANOVA to see the difference between different populations regarding gait parameters)
New diagnostic/ predictive approaches: ML, graph models, population analysis models
Why it is not utilized in real-life applications and/or CDSS:
1. data quality issues [3], [4];
2. Lack of a right and robust model [3];
3. Lack of validated movement feature selection techniques;
4. Lack of model interpretability [3];
5. lack of a sophisticated visualization model for continuous monitoring of individuals’ movement patterns and their associated health conditions over time;
11/18/2019

4. Diagnosis and stage identification Remote monitoring of patients
Introduction
Research Foundations
Research Method
Completed Studies
Conclusion
Research Foundations
Movement Analysis Applications in the Healthcare Domain
Diagnosis and stage identification
Diagnosis and stage identification
Prognosis
Event Detection
Remote monitoring of patients
Remote monitoring of patients
Diagnosis include diagnosis of disease and identifying disease severity level
Prognosis: assess the risk of developing a disease or estimate the effects of an intervention
Event detection: detecting FOG events or Dyskinesia (abnormality or impairment of voluntary movement. Created as a side effect of treatment, mainly levodopa), Bradykinesia (means slowness of movement), tremor (“A tremor is a rhythmic, back-and-forth movement,”)
Remote monitoring
11/18/2019

5. Research Foundations controlled environment vs real-life setting
Introduction
Research Foundations and Goals
Theoretical Founations
Research Method
Completed Studies
Conclusion
Research Foundations
controlled environment vs real-life setting
Lack of a sophisticated visualization model for continuous monitoring
Wrist vs other body locations
More specific gaps
Only 3 studies have been done in free-living conditions. None of them focused on diagnosis purposes for neurodegenerative disease
Although few studies tried to diagnose patients with neurodegenerative disease (e.g., MS and PD) by utilizing gait patterns extracted from the data collected using triaxial sensors located at individuals’ ankle [18], [19] there is no study in the literature on PD diagnosis using the data collected from triaxial sensors on wrist which is the most common and comfortable body location for sensor attachment.
Some may use filtering techniques to get rid of noise [12] and others may ignore the filtering phase and analyze the data the way it is [51], [56], [57]
PD patients and control subjects were not age-matched. Is the difference due to the gait alterations caused by ageing or the disease itself.
Barth et al. has used pressure sensors along with inertial sensors [50
Only one study (Klucken et al.). used inertial sensors including accelerometers and gyroscopes for classifying individuals into healthy and PD groups (92 features)
Lack of a robust model against noise
Lack of an evidence-based feature selection technique for diagnosis purposes
11/18/2019

6. Workflow of Wearable-Based Movement Analysis
Introduction
Research Foundations
Research Method
Completed Studies
Conclusion
Workflow of Wearable-Based Movement Analysis
Target
Data Collection
Preprocessing
Segmentation
Feature Selection and Modelling
Feature Extraction
Workflow of wearable-based gait analysis in the context of diagnostic and preventive healthcare
The goal is to create a robust model:
Data Collection is done using wearable sensors worn by different groups of people:
Parkinson’s, healthy, and elderlies.
Preprocessing
Segmentation.
feature extraction,
Model creation (feature selection and modeling)
Exploratory Data Analysis

7. Body Cites Introduction Research Foundations Research Method
Completed Studies
Conclusion
Body Cites
11/18/2019

8. Data Preprocessing and Segmentation
Introduction
Research Foundations
Research Method
Completed Studies
Conclusion
Data Preprocessing and Segmentation
Segmentation
Step/Stride-based
Peak detection
Dynamic Time Warping
Epoch-based
Noise Filtration: High pass filters, low pas filters, FFT, changing to frequency domain, Butterworth filters
Stride Segmentation
1. Peak detection
2. Dynamic time warping
11/18/2019
Elham Rastegari

9. Document-of-Words Features
Introduction
Introduction
Research Foundations
Research Foundations
Research Method
Research Method
Pilot Studies
Completed Studies
Conclusion
Document-of-Words Features
11/18/2019

10. Feature Selection Proposed Feature Selection:
Introduction
Introduction
Research Foundations
Research Foundations
Research Method
Research Method
Pilot Studies
Conclusion
Feature Selection
Developed for different purposes
1. PCA, LDA, SVD
2. Genetic Algorithm
3. Smallest SD
4.Maximum signal-to-noise ratio (MSN)
5. MSN-Minimum Correlation (MSN_MC)
6. Maximum Prediction Power-MC (MPP-MC)
Proposed Feature Selection:
Maximum Information Gain Minimum Correlation (MIGMC)
relevant features are selected for model development and irrelevant and redundant features are removed to reduce the model complexity and avoid overfitting.
To avoid overfitting, features that lack robustness against sources of variability should be eliminated.
Current algorithms are developed for gait identification or activity classification
Disadvantages of current feature selection techniques in this domain:
PCA and LDA and Singular Value Decomposition: 1) can only produce linear subspace feature extractors. These are unsuitable for highly complex and nonlinear data distributions . 2) not appropriated for diagnosis purposes because of the evidence-based medicine concept
Genetic algorithm: Has not shown a great performance for gait feature selection
Smallest SD: great for authentication purposes
Maximum signal-to-noise ratio
Signal
11/18/2019

11. Maximum Information Gain Minimum Correlation (MIGMC)
Introduction
Introduction
Research Foundations
Research Foundations
Research Method
Research Method
Completed Studies
Completed Studies
Conclusion
Conclusion
Maximum Information Gain Minimum Correlation (MIGMC)
11/18/2019
Elham Rastegari

12. Modeling Similarity Network Model Machine Learning Introduction
Research Foundations
Research Method
Completed Studies
Conclusion
Modeling
Similarity Network Model
Machine Learning
11/18/2019

13. Target Populations Parkinson’s Disease Multiple Sclerosis
Introduction
Research Foundations
Research Method
Completed Studies
Conclusion
Target Populations
Parkinson’s Disease
Multiple Sclerosis
Amyotrophic Lateral Sclerosis
Huntington’s disease
Aging (Geriatrics) …
Healthy Control (age-matched)
11/18/2019
Elham Rastegari

14. Dataset: Participants and Protocol (Ankle Data)
Introduction
Research Foundations
Research Method
Completed Studies
Conclusion
Dataset: Participants and Protocol (Ankle Data)
Protocol:
40 Meter Walking (10-meter walkway back and forth)
Sampling frequency:100
Mild PD
Control PD
Geriatrics
Number of subjects 10
Gender (M/F)
5:5
6:4
Age
64 ± 8.4
63.8 ± 9.3
81 ± 4.1
UPDRS III
12.7 ± 6.0
H & Y
1.7 ± 0.9
11/18/2019
Elham Rastegari

15. Maximum Information Gain Minimum Correlation (MIGMC)
Introduction
Research Foundations
Research Method
Completed Studies
Conclusion
Maximum Information Gain Minimum Correlation (MIGMC) 32 22 8
11/18/2019
Elham Rastegari

16. Selected Set of Features
Introduction
Research Foundations
Research Method
Completed Studies
Conclusion
Selected Set of Features
Feature’s Name
Description
Feature category
Variabiltiy_StrideTime
Variability of stride time
Signal level
Variability_SVM
Variability of vector magnitude
Variability_RMSX
Variability of root mean square in the AP direction
Variability_RMSZ
Variability of root mean square in the ML direction
Velocity
velocity
Smoothness_X
Smoothness in the AP direction
Smoothness_Z
Smoothness in the ML direction
RMSZR
Root mean square relative to the mean value in the ML direction
Stride level
11/18/2019
Elham Rastegari

17. Document of Words Approach
Introduction
Research Foundations
Research Method
Completed Studies
Conclusion
Document of Words Approach
Clustering the subsequence of signals based on a feature set:
SVM, RMSX, RMSY, RMSZ
Slices of 1 sec (Ankle Data)
4 clusters (Words)
11/18/2019
Elham Rastegari

18. Modeling: Machine Learning
Introduction
Research Foundations
Research Method
Completed Studies
Conclusion
Modeling: Machine Learning
Standard Features:
All features (32)
First reduced set of features (22)
Using Information Gain and Ranker methods
Second reduced set of features (8)
Using Pearson Correlation coefficient and ANOVA table
Third reduced set of features (7)
feature sets with one feature less than the optimal number of features
Document-of-Words Features:
10 Features for wrist data and 4 features for ankle data
Various Machine Learning Techniques:
SVM, Random Forest, Naïve Bayes, AdaBoost, and bagging
Validation:
K-Fold Cross validation
Accuracy measures:
F-measure, Precision, Recall
There is always a trade-off between precision and recall. In problems such as diagnosis of PD in the early stages, considering a model with a higher recall value is important because we would not want to miss the diagnosis of patients with PD while misidentification of healthy individuals as PD patients is more tolerable. I should emphasize on the fact that it does not mean that giving a healthy induvial the treatments a PD patient usually receives is acceptable. There is always a need for extra investigations on each individual’s health condition by physicians. Any model created for the healthcare domain is supposed to help physicians, not to replace them. On the other hand, in creating a model for identification of various stages of disease, the precision value is as important as the recall value is. Therefore, we will select a model with high accuracy, and a good balance between precision and recall.
11/18/2019
Elham Rastegari

19. Results-Ankle Data- Standard Features
Introduction
Research Foundations
Research Method
Completed Studies
Conclusion
Results-Ankle Data- Standard Features
Gait Features
SVM
Random Forest
Ada Boost
Bagging
Naïve Bayes
Acc
Pre
Rec
All (32)
Reduced set1(22)
Reduced set2 (8)
Reduced set3 (7)
11/18/2019
Elham Rastegari

20. Results-Ankle Data- Document-of-Words Features
Introduction
Research Foundations
Research Method
Completed Studies
Conclusion
Results-Ankle Data- Document-of-Words Features
Word Features
SVM
Random Forest
Ada Boost
Bagging
Naïve Bayes
Acc
Pre
Rec
4 Words-10 sec
4 Words- 5 sec
4 Words- 2 sec
4-words – 1 sec
Boosting algorithms preform better than others.
Although document-of-Words features and standard features are both showing 100 % performance, the Document of Word approach is using 4 features
11/18/2019
Elham Rastegari

21. Similarity Network Model
Introduction
Research Foundations
Research Method
Completed Studies
Conclusion
Similarity Network Model
Pairwise Correlation
A pairwise Pearson correlation analysis between subjects, using gait parameters
Threshold  85%
Significance  0.05
Cosine Similarity
Considering the Cosine score between a pair of subjects as the measure of similarity
Threshold  20 degrees
Creating Network Model
Vertices represent subjects
If two subjects are highly correlated, there is an edge
11/18/2019
Elham Rastegari

22. Similarity Network Model- Ankle Data (Mild PD) All Features
Introduction
Research Foundations
Research Method
Completed Studies
Conclusion
Similarity Network Model- Ankle Data (Mild PD) All Features
11/18/2019
Elham Rastegari

23. Similarity Network Model-Ankle Data (Mild PD)- Reduced22
Introduction
Research Foundations
Research Method
Completed Studies
Conclusion
Similarity Network Model-Ankle Data (Mild PD)- Reduced22
11/18/2019
Elham Rastegari

24. Similarity Network Model-Ankle Data (Mild PD)- Reduced8
Introduction
Research Foundations
Research Method
Completed Studies
Conclusion
Similarity Network Model-Ankle Data (Mild PD)- Reduced8
11/18/2019
Elham Rastegari

25. Similarity Network Model-Ankle Data (Mild PD)- Reduced7
Introduction
Research Foundations
Research Method
Completed Studies
Conclusion
Similarity Network Model-Ankle Data (Mild PD)- Reduced7
11/18/2019
Elham Rastegari

26. Future Works Analyzing the data from wrist
Introduction
Research Foundations
Research Method
current Studies
Conclusion
Future Works
Analyzing the data from wrist
A longitudinal Similarity Network Approach to assess evolution of disease over time
Comparing various Feature Selection Techniques
Optimization Analysis
Robustness Analysis
11/18/2019

27. Introduction
Research Foundations
Research Method
Completed Studies
Conclusion
Conclusion
Suggesting that using accelerometer data at ankle, it is possible to diagnose neurodegenerative disease such as PD
Emphasizing the significant role of feature selection technique (MIGMC)
Emphasizing the significant role of the Document-of-word model
Emphasizing the significant role of the Similarity Network model
I made the effort to highlight the difference in the pattern of behavior between the two events. The results show that despite the similarity of the broadcasting technology used in the two social events, different patterns of collective behavior emerge.
This pilot study acknowledges the important role of technology in empowering social change, but at the same time suggests that technology is not a restricting factor that dictates the pattern of collective behavior. It would appear that social phenomena play a significant role in shaping the social change.
entangled relationships between technology and social factors.
11/18/2019

28. 11/18/2019
Elham Rastegari

30. Related Publications
Rastegari, Elham, and Hesham Ali. "A Correlation Network Model Utilizing Gait Parameters for Evaluating Health Levels." Proceedings of the 8th ACM International Conference on Bioinformatics, Computational Biology, and Health Informatics. ACM, 2017.
Rastegari, Elham, et al. "Using gait parameters to recognize various stages of Parkinson's disease." Bioinformatics and Biomedicine (BIBM), 2017 IEEE International Conference on. IEEE, 2017.
Rastegari, Elham, et al. “machine learning and similarity network approaches to support automatic classification of Parkinson's disease using accelerometer data” Hawaiian International Conference on System Sciences. HICSS, 2019.
Rastegari, Elham, and Hesham Ali. "A Hierarchical Learning Model for Extracting Public Health Data from Social Media.“ AMCIS, 2017.
Rastegari, Elham, Marmelat Vivien, Hesham Ali. “On Using Gait Parameters to measure the Impact of Aging and Progression of Neurodegenerative Disorders”. (ready to submit)
Rastegari, Elham, and Hesham Ali. “A tutorial on diagnostic and predictive models using body-worn inertial sensors”. (ready to submit)
Rastegari, Elham, and Hesham Ali. “A Document-of-Words Approach in diagnosis of neurodegenerative disease”. (ready to submit)
11/18/2019

31. Publications
Tahmasbi, Nargess, and Elham Rastegari. "A socio-contextual approach in automated detection of cyberbullying." Hawaiian International Conference on System Sciences (2018).
Tahmasbi, Nargess, and Elham Rastegari. "A Socio-Contextual Approach in Automated Detection of Public Cyberbullying on Twitter." ACM Transactions on Social Computing 1.4 (2018): 15.
Azizian, Sasan, et al. "Identifying personal messages: a step towards product/service review and opinion mining." International Conference on Computational Science and Computational Intelligence
Afzali, Farhad, Morrison, Briana, Rastegari, Elham. “Students and Smartphone Usage: Influencing Factors”. ACM CHI Conference on Human Factors in Computing Systems, CHI 2019 (Under Review)
11/18/2019

32. References 11/18/2019 Elham Rastegari
[1]S. D. Kim, N. E. Allen, C. G. Canning, and V. S. Fung, “Postural instability in patients with Parkinson’s disease,” CNS Drugs, vol. 27, no. 2, pp. 97–112, 2013.
[2]J. Jankovic, “Parkinson’s disease: clinical features and diagnosis,” J. Neurol. Neurosurg. Psychiatry, vol. 79, no. 4, pp. 368–376, 2008.
[3]K. R. Chaudhuri, D. G. Healy, and A. H. Schapira, “Non-motor symptoms of Parkinson’s disease: diagnosis and management,” Lancet Neurol., vol. 5, no. 3, pp. 235–245, 2006.
[4]J. M. Hausdorff, “Gait dynamics in Parkinson’s disease: common and distinct behavior among stride length, gait variability, and fractal-like scaling,” Chaos Interdiscip. J. Nonlinear Sci., vol. 19, no. 2, p , 2009.
[5]T. P. Andriacchi and E. J. Alexander, “Studies of human locomotion: past, present and future,” J. Biomech., vol. 33, no. 10, pp. 1217–1224, 2000.
[6]A. Paraschiv-Ionescu, E. E. Buchser, B. Rutschmann, B. Najafi, and K. Aminian, “Ambulatory system for the quantitative and qualitative analysis of gait and posture in chronic pain patients treated with spinal cord stimulation,” Gait Posture, vol. 20, no. 2, pp. 113–125, 2004.
[7]A. Weiss, T Herman, M Plotnik, M Brozgol , “Can an accelerometer enhance the utility of the Timed Up & Go Test when evaluating patients with Parkinson’s disease?,” Med. Eng. Phys., vol. 32, no. 2, pp. 119–125, Mar
[8]B. R. Bloem, J. Haan, A. M. Lagaay, W. van Beek, A. R. Wintzen, and R. A. C. Roos, “Investigation of Gait in Elderly Subjects Over 88 Years of Age,” Top. Geriatr., vol. 5, no. 2, pp. 78–84, Apr
[9]Tripoliti, E.E., Tzallas, A.T., Tsipouras, M.G., Rigas, G., Bougia, P., Leontiou, M., Konitsiotis, S., Chondrogiorgi, M., Tsouli, S. and Fotiadis, D.I., ” Comput. Methods Programs Biomed., vol. 110, no. 1, pp. 12–26, 2013.
[10]N. L. Keijsers, M. W. Horstink, and S. C. Gielen, “Automatic assessment of levodopa-induced dyskinesias in daily life by neural networks,” Mov. Disord., vol. 18, no. 1, pp. 70–80, 2003.
[11]S. Joshi, D. Shenoy, P. L. Rrashmi, K. R. Venugopal, and L. M. Patnaik, “Classification of Alzheimer’s disease and Parkinson’s disease by using machine learning and neural network methods,” in Machine Learning and Computing (ICMLC), 2010 Second International Conference on, 2010, pp. 218–222.
[12]S. Bind, AK Tiwari, AK Sahani, “A survey of machine learning based approaches for Parkinson disease prediction,” Int. J. Comput. Sci. Inf. Technol., vol. 6, no. 2, pp. 1648–1655, 2015.
[13]I. Rustempasic and M. Can, “Diagnosis of parkinson’s disease using fuzzy c-means clustering and pattern recognition,” Southeast Eur. J. Soft Comput., vol. 2, no. 1, 2013.
[14]M. Shahbakhi, D. T. Far, and E. Tahami, “Speech analysis for diagnosis of parkinson’s disease using genetic algorithm and support vector machine,” J. Biomed. Sci. Eng., vol. 7, no. 4, pp. 147–156, 2014.
[15]R. Das, “A comparison of multiple classification methods for diagnosis of Parkinson disease,” Expert Syst. Appl., vol. 37, no. 2, pp. 1568–1572, 2010.
[16]A. Ozcift, “SVM feature selection based rotation forest ensemble classifiers to improve computer-aided diagnosis of Parkinson disease,” J. Med. Syst., vol. 36, no. 4, pp. 2141–2147, 2012.
[17]H.-J. Huppertz, L Möller, M Südmeyer, “Differentiation of neurodegenerative parkinsonian syndromes by volumetric magnetic resonance imaging analysis and support vector machine classification,” Mov. Disord., vol. 31, no. 10, pp. 1506–1517, 2016.
[18]G. Orru, W. Pettersson-Yeo, A. F. Marquand, G. Sartori, and A. Mechelli, “Using support vector machine to identify imaging biomarkers of neurological and psychiatric disease: a critical review,” Neurosci. Biobehav. Rev., vol. 36, no. 4, pp. 1140–1152, 2012.
[19]K. Tejeswinee, G. J. Shomona, and R. Athilakshmi, “Feature Selection Techniques for Prediction of Neuro-Degenerative Disorders: A Case-Study with Alzheimer’s And Parkinson’s Disease,” Procedia Comput. Sci., vol. 115, pp. 188–194, 2017.
[20]A. H. Chen, C. H. Lin, and C. H. Cheng, “New approaches to improve the performance of disease classification using nested–random forest and nested–support vector machine classifiers,” Res. Notes Inf. Sci. RNIS, vol. 14, p. 102, 2013.
[21]C.-W. Cho, W.-H. Chao, S.-H. Lin, and Y.-Y. Chen, “A vision-based analysis system for gait recognition in patients with Parkinson’s disease,” Expert Syst. Appl., vol. 36, no. 3, pp. 7033–7039, 2009.
[22]M. V. Albert, S. Toledo, M. Shapiro, and K. Koerding, “Using mobile phones for activity recognition in Parkinson’s patients,” Front. Neurol., vol. 3, p. 158, 2012.
[23]S. Arora, V Venkataraman, A Zhan, S Donohue, K.M.Biglan, E.R.Dorsey, “Detecting and monitoring the symptoms of Parkinson’s disease using smartphones: A pilot study,” Parkinsonism Relat. Disord., vol. 21, no. 6, pp. 650–653, Jun
[24]S. Patel,  K Lorincz, R Hughes, N Huggins, John Growdon, David Standaert, Metin Akay, Jennifer Dy, Math Welsh, Paolo Bonato,“Monitoring motor fluctuations in patients with Parkinson’s disease using wearable sensors,” IEEE Trans. Inf. Technol. Biomed., vol. 13, no. 6, pp. 864–873, 2009.
11/18/2019
Elham Rastegari

33. References 11/18/2019 Elham Rastegari
[25]M. G. Tsipouras, A. T. Tzallas, G. Rigas, P. Bougia, D. I. Fotiadis, and S. Konitsiotis, “Automated Levodopa-induced dyskinesia assessment,” in Engineering in Medicine and Biology Society (EMBC), 2010 Annual International Conference of the IEEE, 2010, pp. 2411–2414.
[26]G. Rigas et al., “Assessment of tremor activity in the Parkinson’s disease using a set of wearable sensors,” IEEE Trans. Inf. Technol. Biomed., vol. 16, no. 3, pp. 478–487, 2012.
[27]J. Barth, J. Klucken, P. Kugler, T. Kammerer, R. Steidl, J. Winkler, J. Hornegger, B. Eskofier , “Biometric and mobile gait analysis for early diagnosis and therapy monitoring in Parkinson’s disease,” in Engineering in Medicine and Biology Society, EMBC, 2011 Annual International Conference of the IEEE, 2011, pp. 868–871.
[28]J. Klucken,  J. Barth, P. Kugler, J. Schlachetzki, T Henze, F. Marxreiter, Z. Kohl, R. Steidl, J. Harnegger, B. Eskofier, J. Winkler, “Unbiased and Mobile Gait Analysis Detects Motor Impairment in Parkinson’s Disease,” PLOS ONE, vol. 8, no. 2, p. e56956, Feb
[29]J. Barth, Jens Barth, Michael Sünkel, Katharina Bergner, Gerald Schickhuber, Jürgen Winkler, Jochen Klucken, Björn Eskofier, “Combined analysis of sensor data from hand and gait motor function improves automatic recognition of Parkinson’s disease,” in 2012 Annual International Conference of the IEEE Engineering in Medicine and Biology Society, 2012, pp. 5122–5125.
[30]D. Joshi, A. Khajuria, and P. Joshi, “An automatic non-invasive method for Parkinson’s disease classification,” Comput. Methods Programs Biomed., vol. 145, pp. 135–145, 2017.
[31]J. Gubbi, R. Buyya, S. Marusic, and M. Palaniswami, “Internet of Things (IoT): A vision, architectural elements, and future directions,” Future Gener. Comput. Syst., vol. 29, no. 7, pp. 1645–1660, 2013.
[32]M. Chan, D. Estève, C. Escriba, and E. Campo, “A review of smart homes—Present state and future challenges,” Comput. Methods Programs Biomed., vol. 91, no. 1, pp. 55–81, 2008.
[33]H. Alemdar and C. Ersoy, “Wireless sensor networks for healthcare: A survey,” Comput. Netw., vol. 54, no. 15, pp. 2688–2710, 2010.
[34]J. Bart, C Oberndorfer, C Pasluosta, S Schülein, Heiko Gassner, Samuel Reinfelder, Patrick Kugler, Dominik Schuldhaus, Jürgen Winkler, Jochen Klucken, Björn M. Eskofier, “Stride Segmentation during Free Walk Movements Using Multi-Dimensional Subsequence Dynamic Time Warping on Inertial Sensor Data,” Sensors, vol. 15, no. 3, pp. 6419–6440, Mar
[35]A. Burns, BR Greene, MJ McGrath, Terrance J. O'Shea, Benjamin Kuris, Steven M. Ayer, Florin Stroiescu, Victor Cionca, “SHIMMER: A Wireless Sensor Platform for Noninvasive Biomedical Research,” IEEE Sens. J., vol. 10, no. 9, pp. 1527–1534, Sep
[36]U. Lindemann, B Najafi, W Zijlstra, K Hauer, R Muche, C. Becker, K. Aminian, “Distance to achieve steady state walking speed in frail elderly persons,” Gait Posture, vol. 27, no. 1, pp. 91–96, Jan
[37] D. Kobsar, C. Olson, R. Paranjape, T. Hadjistavropoulos, and J. M. Barden, “Evaluation of age-related differences in the stride-to-stride fluctuations, regularity and symmetry of gait using a waist-mounted tri-axial accelerometer,” Gait Posture, vol. 39, no. 1, pp. 553–557, Jan
[38] J. M. Hausdorff, M. E. Cudkowicz, R. Firtion, J. Y. Wei, and A. L. Goldberger, “Gait variability and basal ganglia disorders: Stride-to-stride variations of gait cycle timing in parkinson’s disease and Huntington’s disease,” Mov. Disord., vol. 13, no. 3, pp. 428–437, 1998.
[39] S. Okud, S Takano, M Ueno, Y Hara,  Yasushi Chida, Tomoko Ikkaku, Fumio Kanda, Tatsushi Toda, “Gait analysis of patients with Parkinson’s disease using a portable triaxial accelerometer,” Neurol. Clin. Neurosci., vol. 4, no. 3, pp. 93–97, May 2016.
[40] M. Sekine, T. Tamura, M. Yoshida, Y. Suda, Y. Kimura, H. Miyoshi, Y. Kijima, Y. Higashi and T. Fujimoto, “A gait abnormality measure based on root mean square of trunk acceleration,” J. NeuroEngineering Rehabil., vol. 10, p. 118, Dec
[41]P. Fazio,  G Granieri, I Casetta, E Cesnik, S. Mazzacane, P. Caliandro, F Pedrielli, E Granier, “Gait measures with a triaxial accelerometer among patients with neurological impairment,” Neurol. Sci., vol. 34, no. 4, pp. 435–440, Apr
[42]E. Rastegari and H. Ali, “A Correlation Network Model Utilizing Gait Parameters for Evaluating Health Levels,” in Proceedings of the 8th ACM International Conference on Bioinformatics, Computational Biology, and Health Informatics, 2017, pp. 568–574.
[43]L. A. Talbot, J. M. Gaines, T. N. Huynh, and E. J. Metter, “A Home-Based Pedometer-Driven Walking Program to Increase Physical Activity in Older Adults with Osteoarthritis of the Knee: A Preliminary Study,” J. Am. Geriatr. Soc., vol. 51, no. 3, pp. 387–392, Mar
[44]J. M. Paquet, B. Auvinet, D. Chaleil, and E. Barrey, “[Analysis of gait disorders in Parkinson’s disease assessed with an accelerometer],” Rev. Neurol. (Paris), vol. 159, no. 8–9, pp. 786–789, Sep
11/18/2019
Elham Rastegari

34. References 11/18/2019 Elham Rastegari
[45]E. Rastegari, V. Marmelat, L. Najjar, D. Bastola, and H. H. Ali, “Using gait parameters to recognize various stages of Parkinson,” in 2017 IEEE International Conference on Bioinformatics and Biomedicine (BIBM), 2017, pp. 1647–1651.
[46]G. Yogev, M. Plotnik, C. Peretz, N. Giladi, and J. M. Hausdorff, “Gait asymmetry in patients with Parkinson’s disease and elderly fallers: when does the bilateral coordination of gait require attention?,” Exp. Brain Res., vol. 177, no. 3, pp. 336–346, Mar
[47]B. Galna, S. Lord, and L. Rochester, “Is gait variability reliable in older adults and Parkinson’s disease? Towards an optimal testing protocol,” Gait Posture, vol. 37, no. 4, pp. 580–585, Apr
[48]R. Baltadjieva, N. Giladi, L. Gruendlinger, C. Peretz, and J. M. Hausdorff, “Marked alterations in the gait timing and rhythmicity of patients with de novo Parkinson’s disease,” Eur. J. Neurosci., vol. 24, no. 6, pp. 1815–1820, 2006.
[49]M. Nallegowda, M., Singh, U., Handa, G., Khanna, M., Wadhwa, S., Yadav, S.L., Kumar, G. and Behari, M., “Role of Sensory Input and Muscle Strength in Maintenance of Balance, Gait, and Posture in Parkinson’s Disease: A Pilot Study,” Am. J. Phys. Med. Rehabil., vol. 83, no. 12, p. 898, Dec
[50]S. Volpato and J. M. Guralnik, “The Different Domains of the Comprehensive Geriatric Assessment,” in Comprehensive Geriatric Assessment, Springer, 2018, pp. 11–25.
[51]N. M. Tahir and H. H. Manap, “Parkinson Disease Gait Classification based on Machine Learning Approach,” J. Appl. Sci., vol. 12, no. 2, pp. 180–185, 2012.
[52]L. Bao and S. S. Intille, “Activity recognition from user-annotated acceleration data,” in International Conference on Pervasive Computing, 2004, pp. 1–17.
[53]C. Ding and H. Peng, “Minimum redundancy feature selection from microarray gene expression data,” J. Bioinform. Comput. Biol., vol. 3, no. 02, pp. 185–205, 2005.
[54]A. K. Tiwari, “Machine learning based approaches for prediction of Parkinson disease,” Mach Learn Appl, vol. 3, no. 2, pp. 33–39, 2016.
11/18/2019
Elham Rastegari

35. A Gait Cycle and its Phases and Sub-Phases
11/18/2019

36. Various Gait Patterns
11/18/2019

37. Dataset: Participants and Protocol (Ankle Data)
Introduction
Research Foundations
Research Method
Completed Studies
Conclusion
Dataset: Participants and Protocol (Ankle Data)
Protocol:
4 minute Walking (around the hospital)
Sampling frequency:100
Moderate PD
Control PD
Geriatrics
Number of subjects 5
Gender (M/F)
3:2
2:3
Age
64 ± 10
72 ± 6.3
81 ± 5.9
UPDRS III
20.8 ± 6.1
H & Y
2.6 ± 0.5
11/18/2019
Elham Rastegari

38. Peak Detection Introduction Research Foundations Research Method
Completed Studies
Conclusion
Peak Detection
Stopping Threshold Calculation
1/5 * std (whole Signal)
(SD_X > Tr) & (SD_Y > T) & (SD_Z>=Tr )?
Segmented by 1 second
More than 3 consecutive segments?
Yes No
Moving
Stopping
(SD_X > Tr) & (SD_Y > T) & (SD_Z>=Tr )?
Segmented by 0.1 second
More than 2 consecutive segments?
Removal of non-gait peaks :
Interval between two peaks  if less than 0.4 sec (this threshold depends on the population)
Amplitude of peak  if less than a threshold
Yes No
Swing
Stance Phase
Removal of non-gait peaks
Positive peak detection in X and Y acceleration
11/18/2019

39. Dynamic Time Warping Introduction Research Foundations Research Method
Completed Studies
Conclusion
Dynamic Time Warping
Movement sequence
Data normalization
Calculation of distance matrix
Calculation of accumulated cost matrix
Distance function and start point of warping path
Template generation
Data normalization
Calculation of warping path
a range of [-1; 1].
normalization was done by dividing the signals by the positive values of the sensor range, which we re 6 g for the accelerometer data
Distance Matrix: D(m,n) = sqrt ((tm-ts)2)
Then we add them up  Dx + Dy + Dz
Accumulated matrix:
First bottom row is the same as D
First column  each element is the accumulated value of the previous elements
Remaining  min value of three neighbors
:  top row shows the accumulated cost function and the end of stride
Based on the cost function,we look for the local minima (based on a threshold)
11/18/2019

40. Which Classifier Performs Better?
Introduction
Research Foundations
Research Method
Completed Studies
Conclusion
Which Classifier Performs Better?
The Wilcoxon signed-rank test 
non-parametric statistical hypothesis test 
General Hypothesis:
H0: All classifiers’ performances are equal
Ha: not all classifiers’ performances are equal
Specific Hypothesis:
H0: Classifier A’s performance <= Classifier B’s performance
Ha: Classifier A’s performance > Classifier B’s performance
Reject H0
Kernel SVM
Random Forest
KNN
11/18/2019
Elham Rastegari

41. Stride level and signal level features
Introduction
Research Foundations
Research Method
Completed Studies
Conclusion
Traditional Standard Feature Extraction Approach: Gait Parameters and Measures
Stride level and signal level features
11/18/2019
Elham Rastegari