Gait Detection

Background

Currently, analysis of the IMU data from home assessments consists of two main applications:

1. objective quantification of physical activity in a general sense, and
2. detection of walking bouts, and subsequent quantitative and qualitative clinical gait analysis

1. Objective quantification of physical activity
Physical activity (PA) is quantified by means of the so-called signal magnitude area (SMA). This measure has previously been used as a feature to detect dynamic activities from accelerometer readings, and its main benefit over simply using the acceleration norm is that it accounts for both amplitude and duratoin of a certain activity.
For description of corresponding reference works, see: https://bit.ly/3FaLfDB

2. Detection of walking bouts, and subsequent clinical gait analysis
Walking bouts (WBs) are detected from the mediolateral angular velocity of an ankle-worn inertial measurement unit (IMU). Here, ankle-worn refers to a device that is worn laterally just above the ankle joint.
For detailed description of the methods and corresponding reference works, see: https://bit.ly/3JOMVDN

Code

1. Object quantification of physical activity

To run compute the SMA for any given device from any given study participant, simply run the batch script:

>> batch_SMAs

Within this script, you need to specify some local definitions:

1. root_dir: this is the root directory, i.e., where the dataset is stored (e.g., /mnt/neurogeriatics_data/Braviva/Data/rawdata).

The batch script will then iterate over the participants (sub_ids), determine for each sub_id the available sessions (e.g., ses-T1) and determine for each sessions the available raw data files. Then, for each data file, the number of devices is determined and for each device individually the SMA over the entire time series is calculated. For calculating the SMA, you additionally can specify some epoch time, which is currently set to 60 seconds.

2. epoch: epoch time, in seconds, used for calculating the SMA

Data

Source Data

The Movisens Move 4 source data consist of 24/7 time series data for each of the devices (i.e., low back, wrist, ankle sensor). The data are in binary format and contain sensing modalities for linear acceleration (acc.bin), angular velocity (angularrate.bin), barometric pressure (press.bin) and ambient temperature (temp.bin). Relevant meta-data, such as sampling frequency and units, are available from a accompanying unisens.xml file. In order to get from source data to a raw data format that can be further processed, the sensor data are accummulated on a per-day basis, and both data and meta-data are saved in a MATLAB struct.

Raw Data

The resulting raw data are organized in a BIDS-like format, i.e.:

.
├── BraViva/
│   ├── rawdata/
│   │   ├── sub-COKI10147/
│   │   │   └── ses-T2/
│   │   │       └── motion/
│   │   │           ├── sub-COKI10147_ses-T2_tracksys-imu_date-20191126.mat
│   │   │           ├── sub-COKI10147_ses-T2_tracksys-imu_date-20191127.mat
│   │   │           ├── ...
│   │   │           └── sub-COKI10147_ses-T2_tracksys-imu_date-20191203.mat

The .mat files in the rawdata/ folder contain a single variable, data, namely a MATLAB struct that contains relevant data and corresponding meta-data:

>>> load('./BraViva/rawdata/sub-COKI70030/sess-T2/motion/sub-COKI70030_sess-T2_run-2020028.mat', 'data')
>>> data

tracked point	data	acq time start
ankle	1x4 struct	1x1 datetime
lowBack	1x4 struct	1x1 datetime
wrist	1x4 struct	1x1 datetime

For each of the tracked points, the data collected with the corresponding device is saved in a struct itself:

>>> data(1).data

type	unit	sampling_frequency	data
angularRate	dps	64	Nx3 double
acc	g	64	Nx1 double
press	Pa	8	Nx1 double
temp	Grad Celsius	1	Nx3 double

with $N$ number of timesteps, that is $f_{\mathrm{s}}6060*24$ for a full day of measurement (e.g., 5,529,600 time steps at 64 Hz).

>>> ts_init = data(1).acq_time_start; % initial timestamp
>>> ts = ts_init + seconds((0:size(data(1).data(1).data,1)-1)'/data(1).data(1).sampling_frequency); % timestamps
>>> figure; plot(ts, data(1).data(1).data); % Plots the angular rate data of the ankle sensor

Derived Data

The derived data are saved in a folder structure according to the following proposed structure:

.
├── BraViva/
│   ├── deriveddata/
│   │   ├── sub-COKI10147/
│   │   │   └── ses-T2/
│   │   │       └── sma/
│   │   │           ├── sub-COKI10147_ses-T2_tracksys-imu_trackedpoint-ankle_date-20191126.tsv
│   │   │           ├── sub-COKI10147_ses-T2_tracksys-imu_trackedpoint-ankle_date-20191127.tsv
│   │   │           ├── ...
│   │   │           └── sub-COKI10147_ses-T2_tracksys-imu_trackedpoint-wrist-date_20191203.tsv