Objective: To obtain an accurate method for the detection of activities of daily living (ADL) in subjects with Parkinson’s disease (PD) by using inertial sensors and decision tree classifiers.

Background: Monitoring of performance of subjects with PD during real life is important for the following of interventions. However, this is still a challenge and most of the literature reported limitations related to monitoring during free-living conditions. Artificial Intelligence and wearable sensors constitute a potential alternative due to their accuracy, ubiquitousness, unobtrusiveness, and low cost [1-3].

Method: Five subjects with PD, mean age of 63.5 ± 2.6 years and mean Hoehn-Yahr score of 2. simulated performance of six common ADL while being their on-medication state in the motion lab at the hospital. Performed ADL were: 1) standing, 2) walking, 3) sitting, 4) stair walking, 5) lying and 6) walking holding up a tray with cups. PD subjects wore four inertial sensors on foot, thigh, pelvis, and wrist to register motion while performing ADL in random order. Inertial data was registered at a sampling rate of 250 Hz. Data were preprocessed using a lowpass 2nd order Butterworth filter with a cutoff frequency of 5 Hz and smoothed using a moving average filter with a window size of 6.4 s. The protocol was approved by Institutional Research and ethics committees and subjects signed their informed consent. Binary classification decision trees were build based on 13 input variables using data of inertial sensors were done in MATLAB and validated using leaving out method with 50% of data. Accuracy was calculated and compared across classifiers.

Results: A data matrix containing 179,555 data was used to build and validate three binary classification decision trees: coarse, medium, and fine. Accuracy of classification decision trees varied from 53.8% to 70.6%. The best accuracy was obtained by the fine tree while the worst accuracy was obtained by the coarse tree.

Conclusion: Fine classification decision trees based on inertial sensor data represent a potential alternative for monitoring ADL performed by subjects with PD due to its accuracy. Further research in a broader sample of subjects and activities is needed to investigate the effect on the accuracy of sensors configuration (number and location) and subjects’ medication status.

References: [1] Lonini L, et al (2018). Npj Digit. Med. 1 [2] Nguyen H, et al (2018). IEEE Trans. Neural Syst. Rehabil. Eng. 26 197–204 [3] Rovini E, et al (2017). Front. Neurosci. 11 Researchers thank the Mexican National Council of Science and technology for its support under grant number: CONACYT PEI PROINNOVA 251032

« Back to MDS Virtual Congress 2021

MDS Abstracts - https://www.mdsabstracts.org/abstract/accuracy-in-detection-of-performance-of-activities-of-daily-living-in-subjects-with-parkinsons-disease-by-using-inertial-sensors-and-decision-tree-classifiers/