A machine learning approach to targeted balance rehabilitation in people with Parkinson's disease using a sparse sensor set.

Abstract

Clinical Balance Assessments Often Rely On Functional Tasks As A Proxy For Balance (E.G., Timed Up And Go). In Contrast, Analyses Of Balance In Research Settings Incorporate Quantitative Biomechanical Measurements (E.G., Whole-Body Angular Momentum, H) Using Motion Capture Techniques. Fully Instrumenting Patients In The Clinic Is Not Feasible, And Thus It Is Desirable To Estimate Biomechanical Quantities Related To Balance From Measurements Taken From A Subset Of The Body Segments. Machine Learning Algorithms Are Well-Suited For This Type Of Low- To High-Dimensional Mapping. Thus, Our Objective Was To Develop And Validate An Artificial Neural Network For Estimating Contributions To H From 12 Body Segments Using Only Five Inertial Measurement Units. The Network Was Trained, Tested And Validated On Data From Five Able-Bodied Individuals Performing Forty Trials Each Of A Circuit Involving Complex Walking Tasks, Including Stairs, Ramp, And Direction Changes. The Network Was Also Separately Tested On Four Trials Of An Individual With Parkinson'S Disease Walking On The Circuit. The Output Of The Network Was Strongly Correlated With The Segment Contributions To H In Both Able-Bodied (R= 0.997) And Parkinson'S Disease (R= (0.998) Subjects. The Estimated Values Also Had Low Error Relative To The Signal Magnitude, With The Largest Mean ± SD Rootmean-Squared Errors Of 8.04 ± 1.76% Peak Signal Magnitude In Able-Bodied Individuals And 7.96 ± 0.91% In The Individual With Parkinson'S Disease. These Promising Results Establish The Feasibility Of Using A Sparse Set Of Inertial Measurement Units To Provide Quantitative Data To Clinicians For Targeted Balance Rehabilitation Across Different Patients.