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

Abstract

Parkinson's disease (PD) is a neurodegenerative disease inducing dystrophy of the motor system. Automatic movement analysis systems have potential in improving patient care by enabling personalized and more accurate adjust of treatment. These systems utilize machine learning to classify the movement properties based on the features derived from the signals. Smartphones can provide an inexpensive measurement platform with their built-in sensors for movement assessment. This study compared three feature selection and nine classification methods for identifying PD patients from control subjects based on accelerometer and gyroscope signals measured with a smartphone during a 20-step walking test. Minimum Redundancy Maximum Relevance (mRMR) and sequential feature selection with both forward (SFS) and backward (SBS) propagation directions were used in this study. The number of selected features was narrowed down from 201 to 4-15 features by applying SFS and mRMR methods. From the methods compared in this study, the highest accuracy for individual steps was achieved with SFS (7 features) and Naive Bayes classifier (accuracy 75.3%), and the second highest accuracy with SFS (4 features) and k Nearest neighbours (accuracy 75.1%). Leave-one-subject-out cross-validation was used in the analysis. For the overall classification of each subject, which was based on the majority vote of the classified steps, k Nearest Neighbors provided the most accurate result with an accuracy of 84.5% and an error rate of 15.5%. This study shows the differences in feature selection methods and classifiers and provides generalizations for optimizing methodologies for smartphone-based monitoring of PD patients. The results are promising for further developing the analysis system for longer measurements carried out in free-living conditions.

Conflict of interest statement

Orion Pharma Oy, Suunto Oy, Forciot Oy funded the KÄVELI project. This does not alter our adherence to PLOS ONE policies on sharing data and materials.

Figures

Fig 1. The distribution of PD patients based on the total Unified Parkinson’s Disease Rating Scale (UPDRS).

The modified score is used to evaluate the symptoms, 0 being no symptoms, and 5 denoting full bed rest or using a wheelchair.

Fig 2. General workflow of the machine learning system from [31].

Figure adapted from the original source.

Fig 3. The orientation of the smartphone during 20 step walking tests: X, Y and Z axes denote up-down, right-left and forth-back directions, respectively.

Positive Z axis denotes the walking direction, when walking a straight line.

Fig 4. Example signal from raw ang low-pass filtered walking signal, from the x-axis accelerometer.

Fig 5. One step cycle from a filtered accelerometer signal in the vertical direction (from heel strike to heel strike).

The peaks (green dot) and troughs (red dots) were identified as the local maxima and minima above and below a threshold, respectively. Thresholds are indicated by the green and red dotted lines. Two segments, Ai and Bi were measured for each step i. A denotes the time from the beginning of a heel strike to a toe-off. B denotes the time from the toe-off to the next heel strike.

Fig 6. Accuracy plotted for 5–100 features selected with the mRMR algorithm and using the Support Vector Machine classifier, logistic regression and linear discriminant analysis.

The highest result for each classifier is marked with an x.

Fig 7. Statistical testing results for Cochran’s Q test and post-hoc test with pairwise comparison of individual step classification.

P-values above 0.01 are marked with green background, and the best five classifiers compared are marked with blue background and white text. The classifier names with the highest accuracy are also bolded. The classifier pairs are placed in the order of the p-value.