[Paper Review] Evaluating Performance of Machine Learning Models for Diabetic Sensorimotor Polyneuropathy Severity Classification using Biomechanical Signals during Gait

Fahmida Haque, Mamun Bin Ibne Reaz+11 more|arXiv (Cornell University)|May 21, 2022

Muscle activation and electromyography studies4 citations

TL;DR

This study proposes a machine learning framework to classify diabetic sensorimotor polyneuropathy (DSPN) severity using biomechanical signals from gait, combining EMG (tibialis anterior, vastus lateralis, gastrocnemius medialis) and 3D ground reaction forces (GRF). Using feature selection and ensemble modeling, it achieves 94.78% accuracy in GRF-based classification and 92.89% in EMG-based classification with optimized feature sets.

ABSTRACT

Diabetic sensorimotor polyneuropathy (DSPN) is one of the prevalent forms of neuropathy affected by diabetic patients that involves alterations in biomechanical changes in human gait. In literature, for the last 50 years, researchers are trying to observe the biomechanical changes due to DSPN by studying muscle electromyography (EMG), and ground reaction forces (GRF). However, the literature is contradictory. In such a scenario, we are proposing to use Machine learning techniques to identify DSPN patients by using EMG, and GRF data. We have collected a dataset consists of three lower limb muscles EMG (tibialis anterior (TA), vastus lateralis (VL), gastrocnemius medialis (GM) and 3-dimensional GRF components (GRFx, GRFy, and GRFz). Raw EMG and GRF signals were preprocessed, and a newly proposed feature extraction technique scheme from literature was applied to extract the best features from the signals. The extracted feature list was ranked using Relief feature ranking techniques, and highly correlated features were removed. We have trained different ML models to find out the best-performing model and optimized that model. We trained the optimized ML models for different combinations of muscles and GRF components features, and the performance matrix was evaluated. This study has found ensemble classifier model was performing in identifying DSPN Severity, and we optimized it before training. For EMG analysis, we have found the best accuracy of 92.89% using the Top 14 features for features from GL, VL and TA muscles combined. In the GRF analysis, the model showed 94.78% accuracy by using the Top 15 features for the feature combinations extracted from GRFx, GRFy and GRFz signals. The performance of ML-based DSPN severity classification models, improved significantly, indicating their reliability in DSPN severity classification, for biomechanical data.

Motivation & Objective

• To develop a reliable machine learning-based classification system for diabetic sensorimotor polyneuropathy (DSPN) severity using gait-related biomechanical signals.
• To address contradictory findings in prior literature on DSPN-related biomechanical changes by leveraging data-driven modeling.
• To evaluate and optimize multiple machine learning models for performance in classifying DSPN severity from EMG and GRF signals.
• To identify the most informative features from EMG and GRF data using a novel feature extraction and ranking approach.
• To compare model performance across different combinations of muscle and GRF component features for optimal classification accuracy.

Proposed method

• Collected raw EMG and GRF signals from three lower limb muscles (TA, VL, GM) and three GRF components (GRFx, GRFy, GRFz) during gait.
• Preprocessed signals using standard techniques to reduce noise and artifacts before feature extraction.
• Applied a newly proposed feature extraction scheme from literature to derive meaningful temporal and spectral features from signals.
• Used the Relief feature ranking algorithm to identify and prioritize the most relevant features, removing highly correlated ones.
• Trained and compared multiple machine learning models, including ensemble classifiers, to identify the best-performing model.
• Optimized the top-performing model using hyperparameter tuning and evaluated performance across various feature combinations.

Experimental results

Research questions

• RQ1 Which machine learning model performs best in classifying DSPN severity using EMG and GRF signals during gait?
• RQ2 What is the optimal set of features extracted from EMG and GRF signals that maximizes classification accuracy?
• RQ3 How does the performance of the model vary when using different combinations of muscle and GRF component features?
• RQ4 Can feature selection techniques significantly improve the accuracy of DSPN severity classification models?
• RQ5 What is the highest achievable classification accuracy using EMG and GRF data with optimized models and feature sets?

Key findings

• The ensemble classifier model achieved the highest performance in classifying DSPN severity among all tested models.
• For EMG signals, the best accuracy of 92.89% was achieved using the top 14 features from combined TA, VL, and GM muscle data.
• For GRF signals, the model reached 94.78% accuracy using the top 15 features from GRFx, GRFy, and GRFz components.
• Feature selection via the Relief algorithm significantly improved model performance by eliminating redundant and irrelevant features.
• The optimized ensemble model demonstrated high reliability and robustness in classifying DSPN severity across different feature combinations.
• The study confirms that machine learning models trained on biomechanical gait signals can effectively classify DSPN severity with high accuracy.