eduzhai > Applied Sciences > Engineering >

Machine learning discrimination of Parkinsons Disease stages from walker-mounted sensors data

  • Save

... pages left unread,continue reading

Document pages: 8 pages

Abstract: Clinical methods that assess gait in Parkinson s Disease (PD) are mostlyqualitative. Quantitative methods necessitate costly instrumentation orcumbersome wearable devices, which limits their usability. Only few of thesemethods can discriminate different stages in PD progression. This study appliesmachine learning methods to discriminate six stages of PD. The data wasacquired by low cost walker-mounted sensors in an experiment at a movementdisorders clinic and the PD stages were clinically labeled. A large set offeatures, some unique to this study are extracted and three feature selectionmethods are compared using a multi-class Random Forest (RF) classifier. Thefeature subset selected by the Analysis of Variance (ANOVA) method providedperformance similar to the full feature set: 93 accuracy and had significantlyshorter computation time. Compared to PCA, this method also enabled clinicalinterpretability of the selected features, an essential attribute to healthcareapplications. All selected-feature sets are dominated by information theoreticfeatures and statistical features and offer insights into the characteristicsof gait deterioration in PD. The results indicate a feasibility of machinelearning to accurately classify PD severity stages from kinematic signalsacquired by low-cost, walker-mounted sensors and implies a potential to aidmedical practitioners in the quantitative assessment of PD progression. Thestudy presents a solution to the small and noisy data problem, which is commonin most sensor-based healthcare assessments.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×