Interpreting Stroke-Impaired Electromyography Patterns through Explainable Artificial Intelligence

Abstract

Electromyography (EMG) proves invaluable myoelectric manifestation in identifying neuromuscular alterations resulting from ischemic strokes, serving as a potential marker for diagnostics of gait impairments caused by ischemia. This study aims to develop an interpretable machine learning (ML) framework capable of distinguishing between the myoelectric patterns of stroke patients and those of healthy individuals through Explainable Artificial Intelligence (XAI) techniques. The research included 48 stroke patients (average age 70.6 years, 65% male) undergoing treatment at a rehabilitation center, alongside 75 healthy adults (average age 76.3 years, 32% male) as the control group. EMG signals were recorded from wearable devices positioned on the bicep femoris and lateral gastrocnemius muscles of both lower limbs during indoor ground walking in a gait laboratory. Boosting ML techniques were deployed to identify stroke-related gait impairments using EMG gait features. Furthermore, we employed XAI techniques, such as Shapley Additive Explanations (SHAP), Local Interpretable Model-Agnostic Explanations (LIME), and Anchors to interpret the role of EMG variables in the stroke-prediction models. Among the ML models assessed, the GBoost model demonstrated the highest classification performance (AUROC: 0.94) during cross-validation with the training dataset, and it also overperformed (AUROC: 0.92, accuracy: 85.26%) when evaluated using the testing EMG dataset. Through SHAP and LIME analyses, the study identified that EMG spectral features contributing to distinguishing the stroke group from the control group were associated with the right bicep femoris and lateral gastrocnemius muscles. This interpretable EMG-based stroke prediction model holds promise as an objective tool for predicting post-stroke gait impairments. Its potential application could greatly assist in managing post-stroke rehabilitation by providing reliable EMG biomarkers and address potential gait impairment in individuals recovering from ischemic stroke.

Keywords: Anchors; LIME; SHAP; electromyography; explainable AI; stroke.

Figure 1

Conceptual diagram of an explainable EMG-based of stroke-impaired gait prediction model using XAI approaches. (a) The EMG Channel Description and sample EMG signal. (b) Feature extraction of EMG spectral features. (c) Feature reduction through feature selection approach. (d) Overview of various ML models with sample comparative performance matrices. (e) State-of-the-art explainable AI approaches (LIME, SHAP, Anchors) for interpretation of stroke prediction models.

Figure 2

Performance parameters and Receiver Operating Characteristic (ROC) curves for k-fold (k = 10) cross-validated classification of stroke and healthy control groups using ML models. (a) Violin plot of performance parameters of k-fold cross-validated model for classification of stroke and healthy control groups using ML models. (b) Cross-validated ROC curve for Gradient Boosting (GBoost) Classifier; (c) cross-validated ROC curve for Random Forest (RF) classifier; (d) cross-validated ROC curve for Histogram Gradient Boosting (HistGBoost) Classifier. Area under ROC curve (AUC) is an indicator of prediction accuracy. The diagonal black dotted line is the reference line showing 50% accuracy.

Figure 3

Receiver Operating Characteristic (ROC) curves for classification of stroke and healthy control groups using testing dataset. Area under ROC curve (AUC) is an indicator of prediction accuracy. The diagonal blue dotted line is the reference line showing 50% accuracy.

Figure 4

Performance matrices of ML models for classification of stroke and healthy control groups using test dataset. (a) Accuracy of RF and GBoost models; (b) precision of RF and GBoost models; (c) recall of RF and GBoost models; (d) F1-score of RF and GBoost models; (e) confusion matrix of test dataset for GBoost classifier; (f) confusion matrix of test dataset for RF classifier; (g) confusion matrix of test dataset for HistGBoost classifier.

Figure 5

SHAP plots interpreting the contributions of EMG features in ML models for classification of stroke and healthy control groups. (a) SHAP feature importance plot for GBoost classifier. (b) SHAP summary plot for GBoost classifier. (c) SHAP feature importance plot for Random Forest classifier. (d) SHAP summary plot for Random Forest classifier.

Figure 6

Visualization of the local contribution of EMG features through the LIME approach in classifying a single test instance (predicted class = stroke) using (a) Gradient Boosting (GBoost) classifier, (b) the Random Forest (RF) classifier; (c) Histogram Gradient Boosting (HistGBoost) classifier. The orange marked cells represent the features that contributed most to classifying the stroke.

Figure 7

Visualization of the local contribution of EMG features through the Anchors NLP XAI approach in classifying a single test instance (predicted class = stroke) using Gradient Boosting (GBoost) classifier.