Distinguishing Pathologic Gait in Older Adults Using Instrumented Insoles and Deep Neural Networks

Area of Science:

• Gerontology
• Biomedical Engineering
• Neurology

Background:

• Gait abnormalities are prevalent in older adults, often stemming from aging and various diseases.
• Distinguishing between different causes of gait issues is complex due to overlapping symptoms in this population.

Purpose of the Study:

• To develop and validate a deep neural network (DNN) model for classifying disease-specific gait patterns in older adults.
• To utilize data from commercialized instrumented insoles for gait analysis.

Main Methods:

• A cohort of 150 older adults (≥ 65 years) was studied, including healthy individuals (HI) and patients with Parkinson's disease (PD), spastic hemiplegic gait (SH), normal-pressure hydrocephalus (NPH), and knee osteoarthritis (OA).
• Participants performed the Timed Up and Go Test (TUGT) while wearing instrumented insoles that collected data from accelerometers and pressure sensors.
• A two-stage DNN model was implemented to first identify healthy individuals and then classify specific pathological gait conditions.

Main Results:

• The DNN model achieved high accuracy in classifying gait patterns, with areas under the curve (AUC) ranging from 0.88 (PD) to 0.98 (OA).
• Specific AUC values were 0.96 for HI, 0.88 for PD, 0.98 for OA, 0.96 for SH, and 0.97 for NPH.
• Statistical analysis revealed significant differences in spatiotemporal gait features among the studied groups.

Conclusions:

• The proposed DNN model effectively classifies various gait abnormalities in older adults using data from simple instrumented insoles.
• This approach offers a reliable method for differentiating disease-specific gaits in the elderly population, aiding in diagnosis and management.