Machine learning assisted classification between diabetic polyneuropathy and healthy subjects using plantar pressure...

Area of Science:

• Biomedical Engineering
• Machine Learning in Healthcare
• Diabetic Neuropathy Research

Background:

• Diabetic polyneuropathy increases foot ulcer risk due to impaired sensation and altered plantar pressure.
• Early detection in ambulatory settings is crucial for managing diabetic patient risk factors.
• Existing research often focuses on signal acquisition rather than integrated analysis.

Purpose of the Study:

• To develop a low-complexity machine learning technique for classifying diabetic polyneuropathy.
• To utilize plantar pressure and temperature signals for automated subject classification.
• To enable early detection of diabetic polyneuropathy in ambulatory health monitoring.

Main Methods:

• Feature extraction using Principal Component Analysis (PCA).
• Feature selection via Maximum Relevance Minimum Redundancy (mRMR).
• Binary classification using a k-Nearest Neighbors (k-NN) classifier.

Main Results:

• Achieved a blind test accuracy of 99.58%.
• Reported high sensitivity (99.50%), precision (99.44%), F1-score (99.47%), and AUC (99.56%).
• Demonstrated low resource hardware implementation (81.2 kB memory, 1.31 s latency on ARM v6).

Conclusions:

• The proposed machine learning technique effectively distinguishes between diabetic polyneuropathy and healthy subjects.
• This approach offers a viable solution for early detection in resource-constrained ambulatory settings.
• The findings support the use of plantar pressure and temperature data for automated diabetic foot complication screening.