Objective: To test whether the acceleration frequency of different body parts during gait can reveal characteristic patterns of Parkinson’s Disease (PD). The technical aim was to verify the ability of spectral analysis in combination with machine learning to correctly classify PD versus healthy controls.

Background: Spectral analysis has been used to analyze biological signals due to its ability to provide information about the distribution of energy in the frequency range [1]. Similar to speech patterns, the frequency of body acceleration changes over time [2]. Based on this similarity, we formulated the hypothesis that certain frequency-based approaches like those typically used in speech analysis [3] can be successfully applied to body acceleration signals during dynamic tasks, e.g., walking.

Method: 36 healthy controls and 43 PD patients were asked to perform an overground walking task wearing 10 IMUs (Inertial Motion Units) placed on the feet, shanks, thighs, lumbar region, chest and wrists. The acceleration signals recorded by each IMU were processed to extract a set of features from their power spectra. These features were used by a generative classification algorithm (Universal Background Model-Gaussian Mixture Models, UBM-GMM) to test the discriminative capacity of each sensor location to differentiate between healthy controls and PD subjects. In addition, features extracted from different IMU locations were evaluated to find the best configuration of sensors.

Results: The most discriminative features were found at the low frequency bands of the power spectra. Features extracted from the right thigh provided the highest classification performance, with an accuracy of 0.81 and an Area Under the ROC curve (AUC) of 0.85. The best combination of sensors was obtained with a configuration of five sensors, located on the thighs, lower back, left foot and right arm, yielding an accuracy of 0.93 and an AUC of 0.95.

Conclusion: Our results suggest that spectral analysis on acceleration signals can be successfully employed to detect PD, especially from the thighs and lumbar region. Considering the single-IMU approach, the proposed approach outperforms previous studies, e.g. [4, 5]. By combining the information from different sensors the performance was remarkably similar that in other studies [6, 7], with the advantage of relying on a simple and easy to interpret set of frequency-domain features.

References: [1] O. Henmi et al., “Spectral Analysis of Gait Variability of Stride Interval Time Series: Comparison of Young, Elderly and Parkinson’s Disease Patients,” 2009. Accessed: Jul. 22, 2020.
[2] H. Dubey, J. C. Goldberg, M. Abtahi, L. Mahler, and K. Mankodiya, “EchoWear: Smartwatch Technology for Voice and Speech Treatments of Patients with Parkinson’s Disease,” doi: 10.1145/2811780.2811957.
[3] Moro-Velazquez, L., Gómez-García, JA, Godino-Llorente, JI, Villalba, J, Rusz, Shattuck-Hufnagel, S., J, Dehak, N. (2019) A forced gaussians based methodology for the differential evaluation of Parkinson’s Disease by means of speech processing. Biomedical Signal Processing and Control, 48: 205-220, 2019
[4] H. Abujrida, E. Agu, and K. Pahlavan, “Machine learning-based motor assessment of Parkinson’s disease using postural sway, gait and lifestyle features on crowdsourced smartphone data,” Biomedical Physics and Engineering Express, vol. 6, no. 3, 2020, doi: 10.1088/2057-1976/ab39a8.
[5] M. I. Juutinen et al., “Parkinson’s disease detection from 20-step walking tests using inertial sensors of a smartphone: Machine learning approach based on an observational case-control study,” 2020, doi: 10.1371/journal.pone.0236258.
[6] C. Caramia et al., “IMU-Based Classification of Parkinson’s Disease from Gait: A Sensitivity Analysis on Sensor Location and Feature Selection,” IEEE Journal of Biomedical and Health Informatics, vol. 22, no. 6, pp. 1765–1774, Nov. 2018, doi: 10.1109/JBHI.2018.2865218.
[7] S. Aich and H.-C. Kim, “Auto Detection of Parkinson’s Disease based on Objective Measurement of Gait Parameters using Wearable Sensors,” International Journal of Advanced Science and Technology, vol. 117, pp. 103– 112, Aug. 2018, doi: 10.14257/ijast.2018.117.09.

« Back to 2022 International Congress

MDS Abstracts - https://www.mdsabstracts.org/abstract/detecting-parkinsons-disease-from-body-limb-acceleration-using-machine-learning-and-a-frequency-domain-analysis/