Navigating features: a topologically informed chart of electromyographic features space.

Area of Science:

• Biomedical Engineering
• Signal Processing
• Machine Learning

Background:

• Effective biological signal pattern recognition relies heavily on selecting relevant features.
• High dimensionality and biological variability lead to feature redundancy and challenges in generalizing optimal feature sets across datasets.
• Current methods often struggle with identifying robust and interpretable feature subsets for classification.

Purpose of the Study:

• To develop a principled and interpretable method for selecting relevant features in biological signal analysis.
• To address the challenge of feature generalizability across different datasets and biological variability.
• To identify functional, non-redundant feature groups within electromyographic (EMG) data.

Main Methods:

• Leveraging topological data analysis to create feature space charts.
• Analyzing feature relationships and similarities to identify functional sub-groups.
• Utilizing multiple electromyographic (EMG) datasets from able-bodied subjects during hand and finger contractions as a case study.
• Recommending representative features based on class separability, robustness, and complexity.

Main Results:

• The developed feature charts successfully identified functional groups among 58 state-of-the-art EMG features.
• These feature groups demonstrated generalizability across three distinct forearm EMG datasets.
• The identified groups represent meaningful, non-redundant information, summarizing different regions of the feature space.
• Representative features were recommended for each group, optimizing classification performance.

Conclusions:

• Topological feature charts offer a powerful tool for principled and interpretable feature selection in biological signal processing.
• The method enhances the generalizability of feature sets, overcoming limitations posed by biological variability.
• This approach facilitates the identification of robust and effective features for accurate pattern recognition in EMG and potentially other biological signals.