Table 2.

Studies on tremor.

Study [Ref]	Technology Description	Location	Subjects	Algorithms	Metrics	Activity	Main Results
[17]	IMU unit. Six-axis inertial sensor on index finger of tremor dominant hand	Hospital	35 PD patients and 22 ET	Autoregression process using Yule-walker method and t-tests.	Power spectrum of subsequences, peak frequency	3 tasks each of 10 s, i.e., kinetic, postural and resting tasks	Temporal fluctuation of resting task can differentiate between PD and ET
[22]	4 inertial sensors taped on hands, feet and around the waist	Clinical	7 PD patients	Wilcoxon’s two-tailed rank sum test, bonferroni correction and spearman’s rank correlation coefficient testing	Angular velocity and power spectral density	Two tests. Rest tremor, while sitting at rest patient was reading a text aloud for 45 s. For action tremor a tapping movement performed for 30 s	Application of DBS come forth in a redistribution of power in the tremor and LF band
[23]	Sensors at 6 different positions of subject’s body i.e., right and left wrists (RW and LW), right and left legs (RL and LL), waist and chest	Clinical	18 PD patients and 5 HS	Hidden Markov’s model	Angle between two sensors and LF energy. For tremor severity classification: spectrum entropy, LF and HF energy, ratio of high to total energy and energy from other body segments	DLA’s	(1) Quantifies tremor severity with 87% accuracy (2) Discriminates tremor from other PD symptoms.
[26]	IMU	Hospital	7 PD patients	Least square estimation models	Amplitude of parkinsonian tremor and dominant frequency of parkinsonian tremor	3 tasks. Rest tremor (RT), postural tremor (PT) and action kinetic tremor assessment (KT). Each last for 10 s.	Measured amplitude correlated well with judgement of neurologists (r = 0.98)
[24]	Kinesia affixed finger worn sensors and wrist worn command module	Clinical	60 PD patients	Multiple linear regression model	Peak power frequency of peak power, RMS of angular velocity and RMS of angle	RT assessed for 30 s when participant remain settle with his hands still in lap, PT for 20 s with arms stretched out infront and KT while participant frequently enlarged his arm and touched his nose for 15 s	Quantitative kinematic features are processed and highly correlated to clinicians scores
[28]	Part 1: 3 uni-axial accelerometers on one wrist. In part 2: same as of part 1 also 2 pairs of uni-axial accelerometers (at stemum and upper dominant leg)	Part 1 in lab and part 2 in home	Part 1: 7 patients, part 2: 59 patients and 43 HS	Part 1: FTFT, detect tremor if longer than minimal duration (1.5 s) of dominant frequency with limited BW. Part 2: same as P1 also determine standing vs. sitting based on gravitational vector	Part 1 measured amplitude, dominant frequency duration and BW. Part 2: same as P1 also measured duration of posture of tremor and mean amplitude	In part 1 seated postures recorded at rest and while performing motor activities. In part 2 measured for 24 h while keeping diary	Part 1: Tremor vs. no tremor compared to specialists: SENS > 82%; SPEC > 93%. Part 2: Duration of tremor moderately correlated with UPDRS score for resting tremor ($\varrho$ = 0.66 standing, 0.77 sitting) Intensity of tremor correlated with resting tremor ($\varrho$ = 0.70 standing, 0.75 sitting)
[30]	Part 1: 3 uni-axial gyroscopes near wrist and part 2: two uni-axial gyroscopes near wrist	Hospital	7 PD patients	IIR filter with 3 s windows and autoregression model. Tremor detected if frequency lies between 3.5 and 7.5 Hz and amplitude >0.92. Tremor amplitude estimated from RMS angular velocity	Dominant pole frequency and amplitude	45 min of 17 ADL while videotaped (DBS on and DBS off). In second part 3–5 h moving freely	Tremor vs. no tremor compared: SENS = 99.5%, SPEC = 94.2%. Estimated tremor amplitude from roll axis showed high correlation (r = 0.87) to the UPDRS tremor subscore.
[19]	For EMG, electrodes at belly and ME6000-biosignal monitoring system is used. Tri-axial accelerometers attached to palmar sides of subjects wrists	Hospital	42 patients and 59 HS	K-means algorithm	Kurtosis variable of EMG (K), crossing rate variable of EMG (CR), correlation dimension and recurrence rate of EMG, sample entropy of acceleration (SampEn), coherence variable of EMG and acceleration (Coh)	Subjects asked to hold their elbows at 90° angle for 10–30 s	According to clustering results one cluster contained 90% HC and two other clusters 76% of patients
[21]	Data from gyroscope and accelerometer	Clinical	23 PD patients	To analyze correlation pearson correlation is used	Acceleration vector and rotation rate vector	Wearing iphone on top of hand while sitting on chair and resting both hands on lap atleast for 30 s. Repeated for both hands	Strong correlation (x > 0.7 and p < 0.01) between patients UPDRS score and signal metrics applied to measure signal