Paper IV – Non-parametric time-domain tremor quantification
8.2 Data acquisition
needed to enable it. As always in engineering and science, the level of detail in such a model is defined by its final use. For tremor quantification, it can be argued that modelling the relations between the measured signals and not the underlying biomechanical system is sufficient because dynamically modifying the system behaviour is not attempted in this framework.
Finally, currently available implementation platforms for tremor quan-tification deserve to be mentioned. Dedicated devices, such as PKG R, are popular and attractive from a regulatory point of view, while smart watches and smart phones are gaining on (Senova et al. 2015). The possibility of se-lecting among a number of competing software solutions on the same hard-ware platform is highly appealing to both the clinical and off-clinic end user.
The contributions of this paper stand as follows:
• A complete method for tremor quantification in time domain based on data measured with a standard smart phone and producing tremor amplitude distribution is presented.
• Deficiencies of spectral methods of tremor quantification related to time-varying signal amplitude and frequency are demonstrated on clin-ical data.
• The proposed tremor quantification method is shown to provide a better decision-making ground in comparing therapeutical effect of DBS settings.
The rest of the paper is organized as follows. First, the clinical data sets utilized throughout the paper are presented. Further, spectral ana-lysis is applied to the data sets in Section 8.3 to highlight the difficulties in tremor quantification related to signal non-stationarity and nonlinearity of the underlying biomechanical system. Further, an analytical model-free time-domain approach to tremor signal extraction is explained step-by-step in Section 8.4. The estimated tremor amplitude is characterized by the stationary behavior of a Markov chain designed from data in Section 8.6.
The obtained results are discussed with respect to clinical and technological relevance in Section 8.7 and, finally, Conclusions are drawn.
8.2 Data acquisition
The experimental data used in the present study have also been described and utilized in previous work (Medvedev et al. 2017). These data have been obtained from one PD patient with an implanted DBS system. Data were collected during a DBS programming session at the Uppsala University Hospital on May 30, 2017, under ethical approval obtained from Uppsala
0 10 20 30 40 50 60
−20
−10 0 10 20
Right Left
Time [s]
Acceleration[m/s2 ]
0 10 20 30 40 50 60
−4
−2 0 2 4
Right Left
Time [s]
Angularvelocity[rad/s]
Figure 8.1: Raw accelerometer and gyroscope signals from the trial with DBS off. The colors denote the x- (blue), y- (green) and z-axes (red) of the respective sensor. The black solid lines indicate which data that belongs to the movement with the right and left hands, respectively. Significant tremor was observed visually for the left hand.
Ethical Review Board. A Samsung Galaxy S5 smart phone was used to collect the data. Only signals from the native sensors of the phone were recorded. Linear acceleration and angular velocity were measured by the accelerometer and the gyroscope, respectively, with the MPU-6500 inertial sensor platform manufactured by InvenSense (InvenSense 2017). The local magnetic field was measured by the AK09911C magnetometer manufactured by Asahi Kasei Microdevices (Asahi Kasei Microdevices Corporation 2017).
120 8.2. Data acquisition
0 10 20 30 40 50 60
−20
−10 0 10 20
Right Left
Time [s]
Acceleration[m/s2 ]
0 10 20 30 40 50 60
−4
−2 0 2 4
Right Left
Time [s]
Angularvelocity[rad/s]
Figure 8.2: Raw accelerometer and gyroscope signals from the trial with DBS1 settings. The colors denote the x- (blue), y- (green) and z-axes (red) of the respective sensor. The black solid lines indicate which data that belongs to the movement with the right and left hands, respectively. Note that here the patient performed four repetitions of the movement instead of three. However, only the first three cycles were used for modeling purposes.
The sample rate of all sensors was fs= 100Hz.
Data were collected while the patient performed a predefined task with the smart phone. The instructions for the task were as follows. Pick up the phone from the table, move it to the ear and put then it back to the same position. The patient was asked to perform this task three times with each hand. The same procedure was repeated for three separate DBS settings:
0 10 20 30 40 50 60
−20
−10 0 10 20
Right Left
Time [s]
Acceleration[m/s2 ]
0 10 20 30 40 50 60
−4
−2 0 2 4
Right Left
Time [s]
Angularvelocity[rad/s]
Figure 8.3: Raw accelerometer and gyroscope signals from the trial with DBS2 setting. The colors denote the x- (blue), y- (green) and z-axes (red) of the respective sensor. The black solid lines indicate which data that belongs to the movement with the right and left hands, respectively.
DBS turned completely off (DBS off), the settings that the patient had since before the visit to the hospital (DBS1), and a new settings to be compared with the old one (DBS2). The data collection procedure took less than one minute for each of the DBS settings.
Fig. 8.1–Fig. 8.3 show the raw accelerometer and gyroscope data ob-tained from the smart phone for the three different DBS settings. Two im-portant features can be noticed in the raw data sets: First, the signals are definitely not stationary, with both frequency and amplitude of the signals
122 8.2. Data acquisition
0 5 10 15 20
10−4 10−3 10−2 10−1 100 101 102 103
bΦa(f)[m2 /s4 /Hz]
Left hand
0 5 10 15 20
10−4 10−3 10−2 10−1 100 101 102 103
Right hand
0 5 10 15 20
10−4 10−3 10−2 10−1 100 101 102 103
Frequency, f [Hz]
bΦg(f)[rad2 /s2 /Hz]
0 5 10 15 20
10−4 10−3 10−2 10−1 100 101 102 103
Frequency, f [Hz]
Figure 8.4: PSD estimates of the detrended accelerometer and gyroscope signals with DBS turned off. The colors denote the x- (blue), y- (green) and z-axes (red) of the respective sensor.
varying throughout the record time. Second, differences between Fig. 8.2 and Fig. 8.3 corresponding to DBS on are difficult to discern by ocular inspection. On the contrary, the signals in Fig. 8.1 clearly confirm more prominent tremor with DBS off. The moment when the phone reaches the highest elevation point is also easy to recognize in each of the plots.
0 5 10 15 20 10−4
10−3 10−2 10−1 100 101 102 103
bΦa(f)[m2 /s4 /Hz]
Left hand
0 5 10 15 20
10−4 10−3 10−2 10−1 100 101 102 103
Right hand
0 5 10 15 20
10−4 10−3 10−2 10−1 100 101 102 103
Frequency, f [Hz]
bΦg(f)[rad2 /s2 /Hz]
0 5 10 15 20
10−4 10−3 10−2 10−1 100 101 102 103
Frequency, f [Hz]
Figure 8.5: PSD estimates of the detrended accelerometer and gyroscope signals for the DBS1 setting. The colors denote the x- (blue), y- (green) and z-axes (red) of the respective sensor.