Table 2 Summary of the inertial sensor-based gait quality measures

INI (Parameter-based)

9 gait parameters: 3 spatiotemporal (GD, SL, and PSP) and 6 kinematic (MV, MH, MHD, MAB, MAD, and SRM)

- Determine eigenvalues and eigenvectors based on normative gait using PCA

- Transform individual and normative gait parameters into new PCA-derived coordinate system

- Euclidean norm used to calculate distance between normative and individual gait (i.e., overall deviation)

MGS (Parameter-based)

6 aspects of gait (amplitude, temporal distribution, complexity, symmetry, and regularity), each comprised of a mix of spatiotemporal or signal-based (e.g., skew, kurtosis) gait parameters

- Calculate eigenvalue/eigenvector pairs, keeping those with eigenvalue ≥ 1

- Per each remaining eigenvector, determine correlation of gait parameters. Keep gait parameter from each “aspect” with the highest correlation

- Using reduced parameter set, calculate z-scores of each parameter based on mean and standard deviation of the normative gait set. Standardize these between 0 and 1

- Using standardized deviation scores per parameter, calculate mean “aspect” scores as well as the overall mean deviation

HMM-SM (Signal-based)

Triaxial accelerometer and gyroscope signals from lower-body inertial sensors

- Each participant’s gait data transformed into multi-gait cycle sequences, using a sliding window to iteratively select groups of 10 gait cycles which are subsequently concatenated along the time axis. Similar to that used in our previous work [29]

- Do same for normative dataset (able-bodied gait)

- Train HMM on the normative able-bodied dataset (\({\lambda }_{control}\))

- Train HMM on the participant’s transformed dataset (\({\lambda }_{p}\))

- Compute similarity between participant and normative HMMs,\(S({\lambda }_{p}|\left|{\lambda }_{control}\right)\)

MDP (Signal-based)

Triaxial accelerometer and gyroscope signals from lower-body inertial sensors

- Train self-organizing map (SOM) on the normative data

- For each time point in gait cycle, find best-matching unit in self-organizing map based off Euclidean norm distance

- Overall score equals mean distance across the gait cycle

DTW (Signal-based)

Triaxial accelerometer and gyroscope signals from lower-body inertial sensors

- Compute distance between each participant gait cycle and able-bodied gait cycle using tslearn algorithm for multivariate time series [39]

- Determine the mean distance for all the comparisons to determine overall DTW-based score