Syrine Slim

Syrine Slim

Master's Thesis

Bradykinesia Detection and Severity Estimation using Gait Analysis in Parkinson’s Disease Patients


Hamid Moradi (M. Sc.), Fleischmann Sophie (M. Sc.), Prof. Dr. Björn Eskofier


03 / 2023 – 09 / 2023


Bradykinesia, the slowness of movement, is one of the cardinal symptoms of Parkinson’s disease (PD); next to tremor, rigidity, and postural instability [1]. Bradykinesia is also known to be the most common symptom among PD patients [2], which associates with dopaminergic deficiency [3, 4].

Motor deficiencies in PD patients, including bradykinesia, are most commonly measured using the unified Parkinson’s disease rating scale (UPDRS). However, UPDRS is a subjective measurement obtained by a professional observing the patient for a short time in the clinic, while bradykinesia fluctuates in a considerable number of PD patients. Therefore, UPDRS may not accurately project the severity of bradykinesia in the patients during the short assessment time.

Obtaining more accurate and informative observations of the patient’s health condition assists clinicians in adjusting the dopaminergic treatments accordingly. Hence an objective and continuous bradykinesia scaling system is advocated. Such a system could be featured in the home-monitoring paradigm, allowing patient-individual treatment and care.

Previous studies have strived to classify the presence of bradykinesia and compared its severity to UPDRS scores [5, 6]. Pastorino et al. [6] conducted a study including 24 PD patients and were able to attain an accuracy of 74.4 ±14.9% in predicting bradykinesia severity using UPDRS scores as the ground truth. In both studies [5, 6], the accelerometer sensors were placed on the limbs, trunk, and belt of the patient. Nevertheless, the complexity of the measurement system makes it impractical for patients to use it for continuous monitoring in free-living conditions. Sam`a et al. [4] assessed the bradykinesia severity by analyzing the gait of 12 PD patients through a waist-worn tri-axial accelerometer. They acquired an average accuracy, sensitivity, and specificity of 91.81%, 92.52%, and 89.07%, respectively. Furthermore, they achieved a correlation coefficient of r>0.9 between the severity and the bradykinesia-related subscores of UPDRS.

Wrist-worn sensors were also used in other studies [7, 8] to detect bradykinesia and assess its severity from non-gait kinematic metrics.

While gait is the eminent locomotion aspect in activities of daily living, a lack of research on bradykinesia severity assessment using gait analysis exists [9]. Also, even fewer studies focused on collecting data from foot-worn IMUs [10].

This work aims to detect bradykinesia using foot-worn IMUs, which deliver an expansive collection of temporal and spatial gait parameters compared to other sensor placements. In addition to bradykinesia, we design a severity detection pipeline using regression models according to the UPDRS-bradykinesia sub-scores. The considered bradykinesia subscores are toe-tapping, leg agility, and global spontaneity of movement, also known as body bradykinesia. Moreover, we examine the model’s performance in the bradykinesia subscores related to the upper extremities, such as finger tapping, hand movements, and pronationsupination movement of hands. We tend to extract gait parameters from the acceleration and gyroscope data and afterward train and validate different classifiers based on these features. We will also investigate the frequency and wavelet transform analysis as an alternative and additional features. These features have been implemented in other fields of gait analysis [11, 12, 13].

We will use three datasets containing foot-worn IMU data. The first dataset was recorded and introduced by Marxreiter et al. [14]. The study includes the data of 13 patients performing 2×10 meter walks in the OFF state and after dopamine inductions in 15-minute

The second dataset will be collected in the course of the thesis. Three patients will be shadowed by a medical professional for two days in a free-living condition during their stay in a clinic. The professional will annotate the existence of the bradykinesia every 15 minutes.

The third dataset is the in-clinic walk tests and the UPDRS scores of the FallRiskPD dataset collected at the university hospital Erlangen.


[1] Shobha S Rao, Laura A Hofmann und Amer Shakil. “Parkinson’s disease: diagnosis and treatment”. In: American family physician 74.12 (2006), S. 2046–2054.
[2] Ronald B Postuma u. a. “MDS clinical diagnostic criteria for Parkinson’s disease”. In: Movement disorders 30.12 (2015), S. 1591–1601.
[3] Ruiping Xia und Zhi-Hong Mao. “Progression of motor symptoms in Parkinson’s disease”. In: Neuroscience bulletin 28.1 (2012), S. 39–48.
[4] Albert Sam`a u. a. “Estimating bradykinesia severity in Parkinson’s disease by analysing gait through a waist-worn sensor”. In: Computers in biology and medicine 84 (2017), S. 114–123.
[5] J Cancela u. a. “A comprehensive motor symptom monitoring and management system: the bradykinesia case”. In: 2010 Annual International Conference of the IEEE Engineering in Medicine and Biology. IEEE. 2010, S. 1008–1011.
[6] Matteo Pastorino u. a. “Assessment of bradykinesia in Parkinson’s disease patients through a multi-parametric system”. In: 2011 Annual International Conference of the IEEE Engineering in Medicine and Biology Society. IEEE. 2011, S. 1810–1813.
[7] N Mahadevan u. a. Development of digital biomarkers for resting tremor and bradykinesia using a wrist-worn wearable device. npj Digital Med 2020 Jan 15; 3: 5.
[8] Jeroen GV Habets u. a. “Rapid dynamic naturalistic monitoring of bradykinesia in Parkinson’s disease using a wrist-worn accelerometer”. In: Sensors 21.23 (2021), S. 7876.
[9] Itay Teshuva u. a. “Using wearables to assess bradykinesia and rigidity in patients with Parkinson’s disease: a focused, narrative review of the literature”. In: Journal of Neural Transmission 126.6 (2019), S. 699–710.
[10] Luca Lonini u. a. “Wearable sensors for Parkinson’s disease: which data are worth collecting for training symptom detection models”. In: NPJ digital medicine 1.1 (2018), S. 1–8.
[11] M Sneha Baby, AJ Saji und C Sathish Kumar. “Parkinsons disease classification using wavelet transform based feature extraction of gait data”. In: 2017 International Conference on Circuit, Power and Computing Technologies (ICCPCT). IEEE. 2017, S. 1–6.
[12] Saeid Rahati, Reihaneh Moravejian und Farhad Mohamad Kazemi. “Gait recognition using wavelet transform”. In: Fifth International Conference on Information Technology: New Generations (itng 2008). IEEE. 2008, S. 932–936.
[13] Pierre Barralon, Nicolas Vuillerme und Norbert Noury. “Walk detection with a kinematic sensor: Frequency and wavelet comparison”. In: 2006 International Conference of the IEEE Engineering in Medicine and Biology Society. IEEE. 2006, S. 1711–1714.
[14] Franz Marxreiter u. a. “Sensor-based gait analysis of individualized improvement during apomorphine titration in Parkinson’s disease”. In: Journal of neurology 265.11 (2018), S. 2656–2665.