Identifying neuropathies through time series analysis of postural tests

Area of Science:

• Biomedical Engineering
• Data Science
• Neurology

Background:

• Postural tests are crucial in physical therapy for diagnosing neuropathies, especially in diabetic patients.
• Current diagnostic methods may benefit from objective, data-driven approaches.
• Diabetic neuropathy presents a significant challenge in clinical practice.

Purpose of the Study:

• To develop and validate a time series analysis method for discriminating between healthy, diabetic, and diabetic neuropathic individuals.
• To assess the efficacy of postural test data in classifying these distinct subject groups.
• To explore the potential of machine learning in conjunction with time series analysis for neurological condition diagnosis.

Main Methods:

• Collected time series data from 32 participants (healthy, diabetic, neuropathic) using a Wii Balanced Board under 8 conditions.
• Analyzed variations in the center of pressure (COP) using statistical and machine learning techniques.
• Employed a probabilistic model based on time series characteristics to generate class-specific average curves for classification.

Main Results:

• Achieved a classification performance (F-score) exceeding 98% for distinguishing between the three groups.
• Identified 'eyes open' and 'eyes closed' conditions as particularly effective for discrimination.
• Developed robust classification models for each participant category.

Conclusions:

• The proposed time series analysis method is highly predictive for classifying individuals based on postural test data.
• This data-driven approach can potentially complement or replace traditional questionnaire-based diagnoses.
• Simplification of diagnostic test protocols is feasible with this advanced analytical method.