Table 3.

Papers about gait and TUG analysis.

References	Tech.	Sensors place	Rec. freq	Experimental design	Subjects	Feature extracted	Analysis/classifiers	Classifier performance or findings
Del Din et al., 2016	ACC	Lower back (L5)	50 Hz	10-m walkway	30 PwPD, 30 HC	Mean, SD, variability and asymmetry of stride, stance and swing time, step length, step velocity	ICC; Pearson correlation; t-Test	ICC > 0.9 for mean step time, stance time, step length both for HC and PwPD. ICC > 0.9 for step velocity for HC. Significant difference in step variability between PwPD/HC
Ferrari et al., 2016	ACC, GYRO	Feet	100 Hz	1st study: walking over a treadmill 1.8, 2.7, 3.5 km/h (1 min). 2nd study: walking in a straight line	1st: 11 HC. 2nd: 16 PwPD	Strides number (#), stride length, stride time, stride velocity	ICC; Zero velocity update algorithm, RMS	Study 1: 4.0% RMS of the differences normalized to the mean stride length. Study 2: 2.9% RMS of the differences in % of the mean stride length. ICC > 0.9
Sejdić et al., 2016	ACC	Lower back (L3)	100 Hz	Gait on treadmill: 3 min walking at preferred pace, rested, 3 min walking at a slower speed (−10% from preferred speed)	10 PwPD, 14 HC, 11 patients with neuropathy	Swing and stance time, single and double support (DS) time, HRs of the trunk ACC	t-Test; ANOVA; Mixed models	Differences in: group (p = 0.04) and speed (p = 0.02) for HRs for ACC/motion capture comparison. Magnitudes of HRs 5–10% lower in ACC than motion capture system.
Palmerini et al., 2013	ACC	Lower back	100 Hz	iTUG test 7 m	20 PwPD, 20 HC	Time of: TUG, sit-to-walk, gait, turning, walk-to-sit. RMS of: sit-to-walk, gait, turning, walk-to-sit. Normalized jerk score of the ACC of: sit-to-walk, gait, turning, walk-to-sit. Step time, SD of step time, CV of step time. Phase of step, SD of phase, CV of phase. Phase coordination index (PCI) during gait. HRs of the trunk ACC	ANOVA; ICC; Pearson correlation; LDA, Quadratic discriminant analysis (QDA), Mahalanobis (MC) classifiers	Misclassification rate: 22.5% for LDA, 27.5% for QDA, 37.5% for MC. Correlation of: (i) time TUG with gait and posture subscore (r = 0.6, p = 0.005), rigidity subscore (r = 0.46, p = 0.04) and PIGD subscore (r = 0.51, p = 0.02); (ii) HRs with gait and posture subscore (r = −0.47, p = 0.037); (iii) turning RMS in the VT direction with gait and posture subscore (r = −0.58, p = 0.007) and rigidity subscore (r = −0.47, p = 0.038). ICC > 0.8 for time TUG but no group differences.
Grimpampi et al., 2013	ACC, GYRO	Waist	100 Hz	Walking at a self-selected speed along a 12 m rectilinear pathway	11 PwPD, 13 after stroke	Pitch, roll and yaw angles, walking speed	RMS and correlation coefficient	RMS < 1° for pitch and roll; RMS = 1.3° for yaw for IMU/optoelectronic system comparison. Coeff = 0.8 for correlation
Trojaniello et al., 2015	ACC, GYRO	Lower back (L4/S2)	128 Hz	Walking back / forth for 1 min along a 12 m walkway with the instrumented mat placed 2 m from the starting line (self-selected, comfortable speed)	10 PwPD, 10 HC, 10 hemiparetics, 10 Huntington disease	Gait cycles, mean and SD of stride time, stance time, swing time, step time, gait velocity	Wilcoxon signed-rank test, Friedman test	Stride time, step time, stance and swing duration errors for PwPD were significantly larger than HC
Mariani et al., 2013	ACC, GYRO	Feet	200 Hz	TUG (3 m) and gait at self-selected speed on moderate (2 × 20 m) and long (4 × 50 m) distance, including straight walking and 180° turns	10 PwPD, 10 HC	Stride velocity, stride length, turning angle, path length, swing width, inter-cycle variability	ICC; mean ± SD	acc. ± prec. of 2.8 ± 2.4 cm/s and 1.3 ± 3.0 cm for stride velocity and stride length estimation compared to optical system
Esser et al., 2013	ACC, GYRO	L4	100 Hz	10-m walkway free of obstacles, self-selected walking pace	14 PwPD, 10HC	Cadence, stride length, walking speed, parameters extracted by phase plot analysis (i.e., spread and width of the cloud data points)	ICC; t-Test; Phase plot variability analysis	ICC > 0.9; Difference (p = 0.041) for walking speed between PD/HC. The width of data point is affected by a change in step length, the spread in data point is affected by a change in cadence and step length
Schmidt et al., 2011	StepWatch Activity Monitor	Wrist	Not reported	15 m walking	20 PwPD and multiple sclerosis (MS)	Stride count	Pearson coefficient	r = 1.0 for PwPD correlation between SAM/GaitMat II. r = 0.99 for MS correlation between SAM/GaitMat II
Parisi et al., 2016	ACC, GYRO	Chest, thighs	102.4 Hz	Gait	34 PwPD	Stride time, stance time, DS time, step and stride length, limp, step velocity, thigh rom, cadence, step regularity, symmetry, spectrum power	PCA; NCC, kNN, SVM	Prec., sens. and spec.: 66.48, 31.83, and 88.03%
Salarian et al., 2010	ACC, GYRO	Sternum, forearms, thighs, shanks	200 Hz	TUG (3 m) and iTUG (7 m) including: gait, turning, Sit to Stand (Si2St), Turn-to-sit	12 PwPD (early), 12 HC	Gait: cadence, stance, DS, limp, ROM of shank, thigh and knee, stride length and velocity, peak swing velocity, arm swing pitch and yaw, peak arm swing velocity, arm swing speed asymmetry, peak trunk horizontal and sagittal velocity, ROM of trunk. Turning: peak angular velocity, duration, steps, average and max step time, # double steps. Si2St: peak and average angular velocity, duration, ROM of trunk. Turn-to-sit: duration, ROM of trunk, steps, average and max step time, # double steps	ICC; Wilcoxon rank-sum test	ICC > 0.9 for temporal measures except for limps; ICC = 0.75 for gait measures; ICC = 0.23 for Si2St measures; ICC = 0.67 for turning measures; ICC = 0.50 for turn-to-sit measures. Differences in cadence (p < 0.006), angular velocity of arm-swing (p < 0.005), turning duration (p < 0.023), and time to perform turn-to-sits (p < 0.023) between early PwPD/HC
Weiss et al., 2010	ACC, ECG, GSR, force sensors	Lower back (L3/L5)	256 Hz	TUG	17 PwPD, 15 HC	Mean and SD of: time of TUG, time of Si2St, time of stand-to-sit (St2Si), Range Si2St, Range St2Si, Jerk Si2St, Jerk St2Si. Median and SD of ACC	t-Test; Pearson correlation	TUG duration (p < 0.02), median ACC (p = 0.02), SD ACC (p < 0.004) higher in PwPD than HC. Range and jerk of ACC lower in PwPD than HC (p < 0.006). Jerk Si2St correlated with UPDRS (r = 0.56; p = 0.02) and HY scores (r = 0.49, p = 0.04)
Giuberti et al., 2015	ACC, GYRO	Chest	102.4 Hz	Si2St	24 PwPD	Forwards/backwards/total duration, forwards/backwards/average bending amplitude and bending speed	PCA; NCC, kNN, SVM	Miscalssification rates: 3.7% for UPDRS = 0; 100% for UPDRS = 0.5; 50% for UPDRS = 1; 71.4% for UPDRS = 2; 100% for UPDRS = 2.5; 100% for UPDRS = 3. No data for UPDRS = 2, 3.5, 4
Curtze et al., 2016	ACC, GYRO	Wrists, ankles, lumbar segment, sternum	Not reported	ISAW: to stand still for 30 s, initiate gait with the most affected leg, walk 7 m at comfortable pace, turn 180°, and walk back to the starting location (3 times)	104 PwPD	34 measures of gait and balance into 6 domains: postural sway, initiation of gait, gait arm and trunk movement, gait dynamic stability, turning	Spearman correlation; false discovery rate correction	30 significant associations between gait and balance measures and clinical scales. Turning and gait–pace are most indicative of patient status
Horak et al., 2016					100 PwPD, 21 HC	90 measures of gait and balance into 6 domains: sway area, sway freq, gait speed, gait trunk, gait timing, arm asymmetry	ICC; Pearson correlation; t-test	ICC > 0.75 for 30 features. Gait and postural sway measures not highly correlated. r = −0.62 between gait trunk domain and PIGD subscale. Largest differences PD-ON/HC in gait speed and gait trunk
Weiss et al., 2011	ACC (Mobi8, TMSI)	Lower back	256 Hz	Validation study: 1 min, straight-line walk at a self-selected, comfortable pace inside a long hallway. Gait test: straight-line walk (~25 m × 2). ADL simulation: 500 m walk at comfortable, self-selected speed	22 PwPD, 17 HC	Stride time, stride time variability (validation study only). Dominant freq, amplitude, width (FD), and slope of the main freq of the PSD in the 0.5–to 3.0-Hz band	t-tests 2-tailed; paired t-tests; Pearson coefficients	Width larger, and amplitude and slope smaller in PwPD compared to HC [validation study and ADL simulation (p < 0.02)]. Width decreased, and amplitude and slope increased with anti-Parkinsonian medications (p < 0.007). Significant correlations ACC-derived measures/UPDRS-Gait5
Rahimi et al., 2014	FAB system BioSyn® (ACC, GYRO)	Head, arm, forearm, trunk, pelvis, thigh, shank	100 Hz	Walking, walking turns of 180° and fast walking (3 trials)	11 PwPD	Mean and peak amplitude values of each of 59 joint variable	Change space; Least Absolute Shrinkage Selection Operator (LASSO)	Correctly predicted 5 cases of improvement and 2 cases of worsening after medication
Salarian et al., 2013	GYRO	Shanks	200 Hz	Walking down a 20 m straight hallway both with comfortable speed and fast speed	10 PwPD, 10 HC	Joint angle kinematics including flexion/extension angles at the hip, knee, ankle joints, cadence, % of swing, stance, DS phases	Linear mixed model	Double pendulum model with 2 GYRO on shanks reduces the number of sensing units compared to more complex methods, with relatively small impact on accuracy
Yoneyama et al., 2016	Portable rhythmo-gram (ACC)	Waist	100 Hz	ADL in the community for 24 h with the device attached at all times (including sleeping hours) except when changing clothes or taking a bath	13 mild and 13 severe PwPD, 13 HC, 13 mild and 13 severe MCI/dementia	Features based on: gait ACC, gait variability, gait cycle, number of gait data	Kruskal-Wallis (KW) test with post hoc Steel-Dwass test	The proposed gait measures may deserve to be used for the quantification of disease-specific context-dependent aspects. Gait variability is the lowest in mild PwPD
Parisi et al., 2015	ACC, GYRO	Chest, thighs	102.4 Hz	Comparative investigation of: Leg agility (LA), Si2St, Gait	34 PwPD	LA: angular amplitude and speed, pause, regularity, repetition freq, thigh inclination, angular velocity power spectrum. Si2St: forwards/backwards/total duration, forwards/backwards/average bending amplitude and bending speed. Gait: stride, stance time and DS time, step and stride length, limp, step velocity, thigh ROM, cadence, step regularity, symmetry, spectrum power	PCA; NCC, kNN and SVM	Prec., sens. and spec.: 34.55, 25.17, 84.52% in LA task; 28.00, 25.63, and 77.51 in Si2St task; 66.48, 31.83, and 88.03% in Gait task
Rochester et al., 2010	ACC, biofeedback	Legs, sternum	25 Hz	Single task, dual task, retention of single and dual task over 3 weeks	76 early and 77 late PwPD	Walking speed, step length, step freq	Multiple linear regression models; 2-tailed analysis	Significant training effect for: Single task in speed with cues, step length with/without cues, cadence without cue; Dual task in speed and step length with/without cues. No significant retention effect
Lord et al., 2010	ACC (Vitaport Activity Monitor)	Not specified	Not reported	(i) Single task: standing up from a chair and walking to the kitchen; (ii) dual motor task: specific ADL; (iii) cognitive task; (iv) multi-task.	29 PwPD	Gait speed, interference effect	Moment correlation coefficients, linear regression	For gait speed: within-subject effect during the functional walk (p < 0.001), with gait speed slower during dual and multi-task performance. Higher UPDRS-III scores resulted in a significantly slower walking speed for all conditions. Participants with impaired sustained attention walked more slowly in single and dual motor conditions.
Demonceau et al., 2015	ACC	Lower back (L3/L4)	100 Hz	Walking at the self-selected pace along a 36-m-long track in a wide, clear, and straight hallway	32 PwPD (HY < 2), 32 PwPD (HY = 2/3), 32 HC	Stride length, cadence, regularity index, symmetry index, walking speed and mechanical powers yielded in the cranial-caudal, AP and ML directions	ANOVA, KW, Tukey, Mann- Witney, Wilcoxon, Correlation coefficients; Multivariate regression, ORs	Difference between PwPD groups in regularity index (p = 0.009, OR = 0.98). Significant difference between PwPD groups and HC in symmetry index, speed, stride length, mechanical powers in ML and AP directions. Significant (p < 0.05) but low (r < 0.5) correlation with clinical data. Regularity index and power in ML direction discriminated the 3 groups.
Yoneyama et al., 2013	ACC	Waist	100 Hz	Test 1: walking 200 steps (9 dictated paces). Test 2: stepping on the same spot and intentional side-to-side body sway; normal walking with the device to the left side of the body; walking asymmetrically; jumping forward with both legs (each 10 s). Test 3: collecting ACC data in hospital (10 min) and outside (24 h)	Test 1: 11 HC; Test 2: 1 HC; Test 3: 12 PwPD	Gait cycle, average VT ACC per cycle	Threshold levels criterion	Test 1: 97.2% sens., 97.4% spec., 97.3% acc. Test 2: 100% sens., 100% spec., 100% acc. Test 3: 94.0% sens., 95.7% spec., 95.4% acc.