MATLAB-to-Python translation and performance evaluation of a gait sequence detection algorithm using lower-back IMU sensor data of different clinical cohorts

Abstract-Text (inkl. Referenzen und Bildunterschriften)

Introduction
The increasing prevalence of mobility-limiting diseases, such as Parkinson's disease (PD), poses a serious burden on healthcare systems. Wearable inertial measurement units (IMUs) allow for long-term monitoring of disease progression and could, therefore, be used to track changes in gait. To assess gait, raw IMU data needs to be processed to obtain clinically relevant parameters. While many studies have focused on a single IMU worn on the lower-back, they often used unavailable software to extract data. However, open-source code implementations for extracting clinically relevant parameters are inevitable to allow data extraction at large scale and, in every clinical setting.

Aim
The current study aimed to translate an already validated MATLAB-based algorithm for detecting gait sequences [1] into Python [2], and to test its performance in healthy adults and five mobility-affected cohorts.

Materials & Methods
We analyzed the data from 100 participants each having 2.5 hours of recordings from a lower-back IMU sensor [3]. The outcomes of the algorithm were compared with reference gaits obtained through insole pressure sensors. Then, we calculated various indices, including sensitivity (True positive gaits divided by the sum of True and Negative gaits), using sliding 0.1s windows, as comprehensively discussed in [1].

Results
We presented sensitivities between Python and MATLAB versus reference gaits for various groups in Fig. 1. The mean sensitivity for different clinical cohorts in the Python-based algorithm is [0.85 0.93], while for the MATLAB-based, is [0.85 0.94].

Summary
The performance of the GSD in Python is comparable to the MATLAB one, across all different clinical cohorts. In our view, this marks an important initial step towards the development of an open-source toolbox for the analysis of human motion data.

References
1. Micó, et al. JNER. 20(1), (2023).
2. https://neurogeriatricskiel.github.io/NGMT
3. Mazzà, et al. BMJ O. 11(12), (2021).