Navigating features: a new scientific paper designs a topologically informed chart of electromyographic features space

Biological pattern recognition systems are finding a growing number of applications, such as computer-aided diagnosis for breast cancer, prosthesis control and brain–computer interfaces. But the success of this kind of pattern recognition depends crucially on the selection of relevant features. Across signal and imaging modalities, a large number of features have been proposed, leading to feature redundancy and the need for optimal feature set identification.

In a new paper, out in Journal of the Royal Society Interface, an international team of scientists including ISI Foundation Principal Researcher Giovanni Petri, ISI Researcher Esther Ibáñez-Marcelo and former ISI researcher Angkoon Phinyomark approaches this problem by leveraging topological tools to create charts of features spaces. Focusing on a specific case study, electromyogram (EMG) signals, researchers identify functional groups among 58 state-of-the-art EMG features (by applying a topology-based data analysis method called Mapper), then select representative features for these groups based on three properties: maximum class separability, robustness and complexity.

Researchers find that these groups describe meaningful non-redundant information, succinctly recapitulating information about different regions of feature space. These results support the usefulness of clustering and feature selection-based topological networks for improving both the performance and understanding of EMG-based pattern recognition. Scientists suggest that future research should consider applying the proposed techniques in other EMG-related research problems such as high-density EMG, gait analysis, speech recognition or detecting neuromuscular abnormalities.
Navigating features: a topologically informed chart of electromyographic features space”, Angkoon Phinyomark, Rami N. Khushaba, Esther Ibáñez-Marcelo, Alice Patania, Erik Scheme, Giovanni Petri. Journal of the Royal Society Interface, 6th December 2017