GAIT PATTERN ANALYSIS FOR CLASSIFICATION OF PARKINSON’S DISEASE AND DEMENTIA

Parkinson’s disease (PD) and Dementia (DM) are common longterm
progressive nero-degenerative diseases caused by cognitive impairments. Patients often endure different levels of the motor system disturbance. It has been found that gait provides many measurable indicators of motor disability and impairment which could contribute to the diagnosis and classification of nerodegenerative diseases. Gait analysis using wearable
devices has been applied to analyze the gait disturbance of patients. Based on the 3axis acceleration data and 3-axis angular velocity data collected by a wearable device, instrumental movement unit (IMU) in a freeliving environment, this thesis studies the differences in gait patterns between patients with nerodegenerative diseases and healthy individuals. Starting from signal data preprocessing, each gait cycle will be identified.
Gait features in the temporal domain, spatial domain as well as signal features of angular velocity are extracted from each circle and its subphases and analyzed. We use the classification model to classify patients with neurodegenerative disease and healthy controls.
By performing feature selection, we obtain the best combination of gait features and the best classification model that could distinguish patients from healthy controls.
Similarly, the classification model is applied to classify patients with Parkinson’s disease and Dementia. And through the analysis of gait features, we obtain the differences in gait features of patients with PD, DM and healthy controls.
As the progress of Parkinson’s disease gets deeper, the motor impairments get worse, and the greater risk of falling they will endure which will cause great damage to their health or even risk their lives. Data from IMU could be also used to identify daily life activities such as falls. A fall detection algorithm combing threshold and classification model is proposed. Fall phase segmentation has been performed and signal features are extracted in each subphase. Combining functional principal component scores of fall data together with signal features extracted from each subphase, our algorithm has a very good performance.

帕金森病(Parkinson’s disease) 和痴呆症(Dementia) 是由认知障碍引起的常见
长期进展性神经退行性疾病。病患者经常忍受不同程度的运动系统障碍。步态为
我们提供了许多测量运动障碍和运动功能损伤指标，有助于神经退行性疾病的诊
断和神经退化性疾病之间的分类。因此，使用可穿戴设备的步态分析可以应用于
分析患者的运动障碍。本文基于自由生活环境下通过一种可穿戴设备，惯性测量
装置(Instrumental Movement Unit) 采集的3 轴加速度数据和3 轴角加速度数据，研
究神经退行性疾病患者与健康个体在步态模式上的差异。从信号数据预处理开始，
我们识别了每个步态周期, 提取并分析了时域、空间上的步态特征，以及角速度的
信号特征。我们运用分类模型对病人和健康人群进行分类并进行特征选择，获得
步态特征的最佳组合，得到了区分患者和健康人群的分类模型。同样，分类模型
还被运用在区分帕金森病和痴呆症上，通过对步态特征的分析，我们获得了两种
疾病在步态特征上的区别。
随着帕金森病的进展，运动障碍越来越严重，病人将承受更大的跌倒风险，这
将对他们的健康造成极大的损害甚至危及生命。可穿戴设备的数据还可以实现跌
倒检测等日常生命活动识别。我们结合阈值和分类模型提出了一种跌倒检测算法。
我们对摔倒的不同阶段进分割，并且在每个子阶段中提取信号特征。对摔倒信号
数据进行函数主成分分析并得到函数主成分得分。把提取出的信号特征和函数主
成分得分作为输入变量训练分类模型，使得摔倒检测算法达到最佳性能。

[1] DORSEY E R, ELBAZ A, NICHOLS E, et al. Global, regional, and national burden of Parkinson’sdisease, 1990–2016: a systematic analysis for the Global Burden of Disease Study 2016[J]. The Lancet Neurology, 2018, 17(11): 939953.
[2] PEREZLLORETS, NEGREPAGESL, DAMIER P, et al. Prevalence, determinants, and effecton quality of life of freezing of gait in Parkinson disease[J]. JAMA neurology, 2014, 71(7):884890.
[3] HOEHN M M, YAHR M D. Parkinsonism: onset, progression and mortality.[J]. Neurology,2001.
[4] GOETZ C G, TILLEY B C, SHAFTMAN S R, et al. Movement Disorder Societysponsoredrevision of the Unified Parkinson’s Disease Rating Scale (MDSUPDRS):scale presentation andclinimetric testing results[J]. Movement disorders: official journal of the Movement DisorderSociety, 2008, 23(15): 21292170.
[5] MIRELMAN A, BONATO P, CAMICIOLI R, et al. Gait impairments in Parkinson’s disease[J]. The Lancet Neurology, 2019, 18(7): 697708.
[6] LORD S, GALNA B, ROCHESTER L. Moving forward on gait measurement: toward a morerefined approach[J]. Movement Disorders, 2013, 28(11): 15341543.
[7] CHRISTOPHER BUCKLEY L A. The Role of Movement Analysis in Diagnosing and MonitoringNeurodegenerative Conditions: Insights from Gait and Postural Control[J]. Brain Sciences,2019, 9(2).
[8] CREVECOEUR F, BOLLENS B, DETREMBLEUR C, et al. Towards a “goldstandard”approachto address the presence of longrangeautocorrelationin physiological time series[J].Journal of Neuroscience Methods, 2010, 192(1): 163172.
[9] HAUSDORFF J M. Gait dynamics in Parkinsondisease: Common and distinct behavior amongstride length, gait variability, and fractallikescaling[J]. Chaos: An Interdisciplinary Journal ofNonlinear Science, 2009, 19(2): 026113.
[10] DEL DIN S, HICKEY A, LADHA C, et al. Instrumented gait assessment with a single wearable:an introductory tutorial[J]. F1000Research, 2016, 5(2323): 2323.
[11] ZIJLSTRA W, HOF A L. Assessment of spatiotemporalgait parameters from trunk accelerationsduring human walking[J]. Gait & posture, 2003, 18(2): 110.
[12] TROJANIELLO D, CEREATTI A, DELLA CROCE U. Accuracy, sensitivity and robustnessof five different methods for the estimation of gait temporal parameters using a single inertialsensor mounted on the lower trunk[J]. Gait & posture, 2014, 40(4): 487492.
[13] GAGE J R. An overview of normal walking[J]. Instructional Course Lectures, 1990, 39: 291303.
[14] PhysioSunit[EB/OL]. https://physiosunit.com/gaitdefinitionphasesofcycleexplained/.
[15] REHMAN R Z U, DEL DIN S, GUAN Y, et al. Selecting clinically relevant gait characteristicsfor classification of early parkinson’s disease: A comprehensive machine learning approach[J].Scientific reports, 2019, 9(1): 112.
[16] LORD S, GALNA B, VERGHESE J, et al. Independent domains of gait in older adults andassociated motor and nonmotor attributes: validation of a factor analysis approach[J]. Journalsof Gerontology Series A: Biomedical Sciences and Medical Sciences, 2013, 68(7): 820827.
[17] DEL DIN S, HICKEY A, HURWITZ N, et al. Measuring gait with an accelerometerbasedwearable: influence of device location, testing protocol and age[J]. Physiological Measurement,2016, 37(10): 1785.
[18] DEL DIN S, ELSHEHABI M, GALNA B, et al. Gait analysis with wearables predicts conversionto parkinson disease[J]. Annals of Neurology, 2019, 86(3): 357367.
[19] GHORAANI B, BOETTCHER L N, HSSAYENI M D, et al. Detection of mild cognitive impairmentand Alzheimer’s disease using dualtaskgait assessments and machine learning[J].Biomedical Signal Processing and Control, 2021, 64: 102249.
[20] WAHID F, BEGG R K, HASS C J, et al. Classification of Parkinson’s disease gait using spatialtemporalgait features[J]. IEEE Journal of Biomedical and Health Informatics, 2015, 19(6):17941802.
[21] TAHIR N M, MANAP H H. Parkinson disease gait classification based on machine learningapproach[J]. Journal of Applied Sciences(Faisalabad), 2012, 12(2): 180185.
[22] PRADHAN C, WUEHR M, AKRAMI F, et al. Automated classification of neurological disordersof gait using spatiotemporalgait parameters[J]. Journal of Electromyography and Kinesiology,2015, 25(2): 413422.
[23] REHMAN R Z U, KLOCKE P, HRYNIV S, et al. Turning detection during gait: Algorithmvalidation and influence of sensor location and turning characteristics in the classification ofparkinson’s disease[J]. Sensors, 2020, 20(18): 5377.
[24] OEDA T, UMEMURA A, TOMITA S, et al. Clinical s associated with abnormal postures inParkinson’s disease[J]. PLoS One, 2013, 8(9): e73547.
[25] ZANARDI A P J, DA SILVA E S, COSTA R R, et al. Gait parameters of Parkinson’s diseasecompared with healthy controls: a systematic review and metaanalysis[J]. Scientific reports,2021, 11(1): 113.
[26] DRANCA L, DE MENDAROZKETA L D A R, GOÑI A, et al. Using Kinect to classify Parkinson’sdisease stages related to severity of gait impairment[J]. BMC bioinformatics, 2018, 19(1): 115.
[27] PHAM M H, ELSHEHABI M, HAERTNER L, et al. Algorithm for turning detection andanalysis validated under homelikeconditions in patients with Parkinson’s disease and olderadults using a 6 degreeoffreedominertial measurement unit at the lower back[J]. Frontiers inNeurology, 2017: 135.
[28] KARLSSON M K, MAGNUSSON H, VON SCHEWELOV T, et al. Prevention of falls in theelderly—a review[J]. Osteoporosis international, 2013, 24(3): 747762.
[29] CHEN J, KWONG K, CHANG D, et al. Wearable sensors for reliable fall detection[C]//2005IEEE Engineering in Medicine and Biology 27th Annual Conference. IEEE, 2006: 35513554.
[30] BOURKE A K, LYONS G M. A thresholdbasedfalldetectionalgorithm using a biaxialgyroscopesensor[J]. Medical engineering & physics, 2008, 30(1): 8490.
[31] WU F, ZHAO H, ZHAO Y, et al. Development of a wearablesensorbasedfall detection system[J]. International Journal of Telemedicine and Applications, 2015, 2015: 576364.
[32] ABBATE S, AVVENUTI M, COLA G, et al. Recognition of false alarms in fall detection systems[C]//2011 IEEE Consumer Communications and Networking Conference (CCNC). IEEE,2011: 2328.
[33] ÖZDEMIR A T, BARSHAN B. Detecting falls with wearable sensors using machine learningtechniques[J]. Sensors, 2014, 14(6): 1069110708.
[34] BECKER C, SCHWICKERT L, MELLONE S, et al. Proposal for a multiphase fall modelbased on realworldfall recordings with bodyfixedsensors[J]. Zeitschrift für Gerontologieund Geriatrie, 2012, 45(8): 707715.
[35] HSIEH C Y, LIU K C, HUANG C N, et al. Novel hierarchical fall detection algorithm using amultiphase fall model[J]. Sensors, 2017, 17(2): 307.
[36] Axivity[EB/OL]. https://axivity.com/product/ax6.
[37] MILLECAMPS A, LOWRY K A, BRACH J S, et al. Understanding the effects of preprocessingon extracted signal features from gait accelerometry signals[J]. Computers in Biologyand Medicine, 2015, 62: 164174.
[38] Bout indentification[EB/OL]. https://matt002.github.io/GaitPy/html/gaitpy_functions.html.
[39] WACHOWIAK M P, RASH G S, QUESADA P M, et al. Waveletbasednoise removal forbiomechanical signals: A comparative study[J]. IEEE Transactions on Biomedical Engineering,2000, 47(3): 360368.
[40] MCCAMLEY J, DONATI M, GRIMPAMPI E, et al. An enhanced estimate of initial contactand final contact instants of time using lower trunk inertial sensor data[J]. Gait & posture, 2012,36(2): 316318.
[41] HSIAO E T, ROBINOVITCH S N. Common protective movements govern unexpected fallsfrom standing height[J]. Journal of Biomechanics, 1997, 31(1): 19.
[42] BIDARGADDI N, KLINGBEIL L, SARELA A, et al. Wavelet based approach for posture transitionestimation using a waist worn accelerometer[C]//2007 29th Annual International Conferenceof the IEEE Engineering in Medicine and Biology Society. IEEE, 2007: 18841887.