Comparison of head-mounted IMU-based methods for temporal gait description: performance on healthy subjects of different age range and parkinsonian patients

IMUs are increasingly popular devices in gait analysis. Various placements of this type of sensor have already been explored in the literature, both in a multi-sensor and single-sensor approach. The possibility of producing these devices on a miniaturised scale has allowed them to be easily integrated into other devices such as smartphones, headphones and smart glasses. The objective of this study was to identify a method that would allow the best possible performance in detecting gait events by analysing the head acceleration of different subjects, belonging to different populations and under different speed conditions. The performance obtained by the newly proposed method, in particular, demonstrates that this objective is achievable. Speaking about the performance obtained by the best trade-off method, on young healthy subjects, the median absolute error ranges from 20 to 10 ms in ICs detection and from 100 to 10 ms in FCs detection with an F1-score always higher than 92%. Performances on the population composed by elderly healthy adults were characterized by mean absolute errors ranging from 50 to 20 ms in ICs detection task and from 90 to 45 in FCs detection task. On Parkinson’s diseased patients the mean absolute error on ICs detection had values fitting in 30 to 20 ms interval while on Fcs the mean absolute error ranged from 60 to 40 ms. In general, the results obtained on ICs recognition task were better than those obtained for FCs detection task among all populations analyzed and better performance were achieved at higher walking speed.