Anselm Hochschild

  • Job title: Master's Thesis
  • Working group: Optimizing HMMs for stride segmentation and event detection

Arne Küderle, M. Sc.,

01/2020 – 06/2020


Parkinson’s disease (PD) is a chronic degenerative disorder of the central nervous system that mainly affects the motor system. PD affects mostly elderly people and an increasing prevalence can be expected in an aging population [1]. Monitoring the symptoms of PD can help to track the disease progression, and hence help to improve patient care [2]. In the past years, various methods have been developed to objectively measure the severity of motor symptoms, especially the impairment of gait. It is likely that the Health related Quality of Life can be improved by technology-based, clinically effective, measurement tools [3]. Objective methods to measure gait include camera based systems, force plates and IMUs [4]. IMUs measure 3D acceleration, and angular velocity. These sensors can be placed on the feet or lower body of patients and can be used in everyday life [5]. A typical gait analysis pipeline, using IMU based data, consists of stride segmentation, followed by extraction of gait parameter (e.g. stride length) [6, 7]. The final spatial-temporal gait parameters can then be used to predict disease stages or monitor the rehabilitation progress of a patient [3]. Stride segmentation and gait event detection have been performed with various approaches, including peak detection, Dynamic Time Warping, and in recent years, Hidden Markov Models (HMM) [8–11]. HMMs deliver promising results as compared to peak detection and Dynamic Time Warping in certain scenarios in the context of gait analysis [10]. One advantage is the possibility to segment gait, and detect relevant biomechanical events in a single analysis step. Yet, previous works did not optimize different HMM structures, regarding structure of the state model, and did not investigate the influence of different features and input data on the behavior and the overall performance of the model. Therefore, further optimization of the method would likely allow to improve the performance of HMMs for gait analysis even further. This work aims to address this issue by implementing different variations of HMMs, and compare them against each other and published methods for stride segmentation. Thereby, not only the architectures, but also input data and features as well as parameters of a model will be varied. Furthermore, this thesis will investigate the influence of certain features on the different parts of the HMM, in particular the underlying probabilistic model for the state emission probabilities [12]. All models will be initially developed and tested on lab data from healthy subjects. The best performing models will be further tested on more challenging datasets from PD patients and/or home monitoring recordings. If successful, these models will be implemented and evaluated against an existing gait pipeline.



[1] Ole-Bjørn Tysnes, Anette Storstein: Epidemiology of Parkinson’s disease Journal of Neural Transmission, 124.8, 2017
[2] Morris ME, Iansek R, Matyas TA, Summers JJ: The pathogenesis of gait hypokinesia in Parkinson’s disease. Brain, 117(Pt 5), 1994
[3] Walter Maetzler, Jochen Klucken, Malcolm Horne: A clinical view on the development of technology-based tools in managing Parkinson’s disease. Movement Disorders 31.9, 2016
[4] Alvaro Muro-de-la-Herran, Begonya Garcia-Zapirain, and Amaia Mendez-Zorrilla Gait Analysis Methods: An Overview of Wearable and Non-Wearable Systems, Highlighting Clinical Applications. Sensors (Switzerland), 14.2, 2014
[5] J. Klucken, J. Barth, P. Kugler, J. Schlachetzki, T. Henze et al.: Unbiased and Mobile Gait Analysis Detects Motor Impairment in Parkinson’s Disease. PLoS ONE, 8.2, 2013
[6] Benoit Mariani, Mayté Castro Jiménez, François J. G. Vingerhoets, Kamiar Aminian: OnShoe Wearable Sensors for Gait and Turning Assessment of Patients With Parkinson’s Disease, IEEE Transactions on Biomedical Engineering, 60.1, 2013
[7] Alexander Rampp, Jens Barth, Samuel Schulein, Karl-Günter Gaßmann, Jochen Klucken, and Björn M. Eskofier: Inertial Sensor-Based Stride Parameter Calculation From Gait Sequences in Geriatric Patients. IEEE Transactions on Biomedical Engineering, 62.4, 2015
[8] Andrea Mannini, Angelo Maria Sabatini: A Hidden Markov Model-Based Technique for Gait Segmentation Using a Foot-Mounted Gyroscope. Annual International Conference of the IEEE Engineering in Medicine and Biology Society, 2011 [9] Andrea Mannini, Diana Trojaniello, Ugo Della Croce, Angelo M. Sabatini: Hidden Markov model-based strategy for gait segmentation using inertial sensors: Application to elderly, hemiparetic patients and Huntington’s disease patients. 37th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC), 2015
[10] Nooshin Haji Ghassemi, Julius Hannink, Christine F. Martindale, Heiko Gaßner, Meinard Müller, Jochen Klucken and Björn M. Eskofier: Segmentation of Gait Sequences in SensorBased Movement Analysis: A Comparison of Methods in Parkinson’s Disease. Sensors (Switzerland), 18.1, 2018
[11] Christine F. Martindale, Florian Hoenig, Christina Strohrmann and Bjoern M. Eskofier: Smart Annotation of Cyclic Data Using Hierarchical Hidden Markov Models. Sensors (Switzerland), 17.10, 2017
[12] Lawrence R. Rabiner: A Tutorial on Hidden Markov Models and Selected Applications in Speech Recognition. Proceedings of the IEEE 77, 1989