Dataset Open Access

IMU data captured unobtrusively and in-the-wild by Parkinson's disease patients and healthy controls

Alexandros Papadopoulos; Konstantinos Kyritsis; Sevasti Bostantjopoulou; Lisa Klingelhoefer; Kallol Ray Chaudhuri; Anastasios Delopoulos

Related person(s)
Heinz Reichmann; Zoe Katsarou; Dimitrios Iakovakis; Dhaval Trivedi; Ioannis Dagklis; Fotis Karayannis; George Ntakakis


The dataset contains IMU signals captured in-the-wild via the accelerometer sensor embedded in modern smartphones, for the purpose of detecting tremorous episodes, related to Parkinson's Disease (PD). A group of 31 PD patients and 14 Healthy controls contributed accelerometer data using their personal smartphones, for a period spanning many months.Tri-axial acceleration values were recorded automatically whenevera phone call was realized. The recording lasted for 75 seconds at the most. Each phone call thus resulted in one recorded accelerometer signal, also referred to as session. Each subject contributed a different amount of sessions depending on the number of phone calls they realized during the data collection period as well as their participation time (they were free to drop-out at any time). A detailed description of the capturing process as well as analysis results, can be found in the related research article.

The data is presented as a list of python dictionaries, stored in a pickle file. Each dictionary in the list, corresponds to one subject and containes the following fields:

1. subject_id: scalar
 A numerical value that uniquely identifies the subject.

2. subject_sessions: list of numpy.array
A list of numpy arrays of shape (N, 4) that contains the tri-axial accelerometer sessions that the subject contributed. N denotes the total length of the session in samples (which varies from session to session) Column 0 of the array contains the timestamps of the accelerometer samples. Columns 1-3 contain the acceleration values across the x,y,z directions.

3. session_datetimes: list of datetime objects 
  A list of datetime objects that denote the capturing date and time of the   corresponding entries in the subject_sessions field.

4. annotation: dict
 A dictionary containing the following tremor-related annotation values:
* updrs16: scalar int
The value related to tremor as described in item 16 of the part II of the MDS-UPDRS scale, as reported by the subject.

* updrs20_right: scalar int in range [0, 4]
The value related to rest tremor in the right hand as described in item 20 of the part III of the MDS-UPDRS scale, as reported by the attending neurologist.

* updrs20_left: scalar int in range [0, 4]
Same as above but for left hand.

* updrs21_right: scalar int in range [0, 4]
The value related to action/postural tremor in the right hand as described in item 21 of the part III of the MDS-UPDRS scale, as reported by the attending neurologist.

* updrs21_left: scalar int in range [0, 4]
Same as above but for left hand.

* sp_expert: scalar int in range [0, 1]
A binary tremor annotation created by a group of signal processing experts, upon visually examining the contributed signals in both time and frequency domain and taking into consideration the UDPRS scores of each subject. This was necessary due to the intermittent nature of tremor, as well as a number of considerations related to the in-the-wild nature of the data capturing process. For more details, we refer the reader to the dataset description in the related research article.
A '1' value indicates that the subject has tremor.
A '0' value indicates that the subject doesn't have tremor.

* pd_status: scalar int in range [0, 1]
A '1' value indicates that the subject is a PD patient.
A '0' value indicates that the subject is a Healthy Control

Note: Each annotation value refers to the subject as a whole, and not in any one session.


The study during which the present dataset was collected is a multi-center study approved in each country available (for more info visit: Informed consent, including permission for third-party access to pseudo-anonymised data, was obtained from all subjects prior to their engagement with the study. The work has received funding from the European Union's Horizon 2020 research and innovation programme under Grant Agreement No 690494 - i-PROGNOSIS: Intelligent Parkinson early detection guiding novel supportive interventions (



Any inquiries regarding this dataset should be adressed to:

Mr. Alexandros Papadopoulos (Electrical & Computer Engineer, PhD candidate)

Multimedia Understanding Groupmug
Department of Electrical & Computer Engineering
Aristotle University of Thessaloniki
University Campus, Building C, 3rd floor
Thessaloniki, Greece, GR54124

Tel: +30 2310 996359, 996365 
Fax: +30 2310 996398



This is an open access dataset, licensed under Creative Commons Attribution 4.0 International (



This dataset comes without any warranty. Administrators of this dataset can not be held accountable for any damage (physical, financial or otherwise) caused by the use of this dataset. 

Files (1.9 GB)
Name Size
1.9 GB Download
  • Papadopoulos, A., Kyritsis, K., Klingelhoefer, L., Bostanjopoulou, S., Chaudhuri, K. R., & Delopoulos, A. (2019). Detecting Parkinsonian Tremor from IMU Data Collected In-The-Wild using Deep Multiple-Instance Learning. IEEE Journal of Biomedical and Health Informatics.

All versions This version
Views 1,6391,639
Downloads 485485
Data volume 915.2 GB915.2 GB
Unique views 1,4301,430
Unique downloads 334334


Cite as