Predicting diabetic peripheral neuropathy through advanced plantar pressure analysis: a machine learning approach

Area of Science:

• Biomedical Engineering
• Diabetology
• Computational Medicine

Background:

• Diabetic foot ulceration (DFU) is a severe complication of diabetes, often leading to amputation.
• Neuropathy in diabetic patients causes insensate feet, increasing plantar tissue stress and ulcer risk.
• Early detection of diabetic peripheral neuropathy is crucial for preventing DFUs.

Purpose of the Study:

• To present a novel plantar pressure distribution analysis method for predicting diabetic peripheral neuropathy.
• To evaluate the efficacy of static versus dynamic plantar pressure assessments.
• To develop an AI-driven system for neuropathy detection and foot classification.

Main Methods:

• Collected clinical and plantar pressure data from 86 diabetic patients using the Win-Track platform.
• Employed an automated image processing algorithm to segment plantar pressure images into forefoot and hindfoot regions.
• Utilized machine learning models (Gradient Boosting, Random Forest, Decision Tree) and explainable AI techniques (SHAP, Eli5, Anchor Explanations).

Main Results:

• Static plantar pressure analysis demonstrated superior performance over dynamic methods, achieving 100% accuracy in predicting neuropathy.
• Gradient Boosting model showed the highest accuracy (88% dynamic, 100% static).
• A foot classification system based on forefoot-hindfoot pressure ratio was developed, categorizing feet as flat, regular, or arched.

Conclusions:

• The developed plantar pressure analysis method shows significant potential for early diabetic peripheral neuropathy detection.
• Explainable AI enhanced model interpretability, providing insights into feature importance.
• This approach can aid in risk stratification, DFU prevention, and reducing amputation rates in diabetic patients.