Frequency-based deep learning to identify subtle postural instability in early, untreated Parkinson's disease

Area of Science:

• Neurology
• Biomedical Engineering
• Data Science

Background:

• Postural instability is a late-stage symptom of Parkinson's disease (PD), often due to insensitive diagnostic measures.
• Early neurodegeneration in PD suggests potential for earlier detection of motor impairments.

Purpose of the Study:

• To develop a sensitive, objective, and easily obtainable measure for early clinical detection of postural instability in Parkinson's disease.
• To utilize machine learning to analyze postural sway signals for distinguishing early PD.

Main Methods:

• Assessed postural sway in 40 newly diagnosed, untreated PD individuals and 79 healthy controls using lumbar spine accelerometry.
• Recorded quiet standing balance for 30 seconds with eyes open and feet together.
• Trained a convolutional neural network (CNN) on sway signal frequency data to differentiate between PD patients and controls.

Main Results:

• The CNN model achieved high diagnostic performance, with an average accuracy of 98.9%, sensitivity of 97.7%, and specificity of 98.9%.
• Characteristic frequency features in postural sway signals effectively identified subtle balance impairments in early PD.
• The method demonstrated significant potential for early and accurate detection of Parkinson's disease.

Conclusions:

• Subtle postural sway impairments, detectable via frequency analysis, are present in early Parkinson's disease.
• This objective measure offers a promising tool for earlier clinical detection and management of PD.
• The findings highlight the potential for integrating advanced signal processing and AI in neurological diagnostics.