Topological Data Analysis for Scalp EEG Signal Processing

Topological Data Analysis is a fast-growing and promising approach that recently gains popularity in the data science field. It utilizes topological and geometric measurements to describe the structure, for example the shape, of complex data, which is fundamental and important for modeling the data. Scalp Electroencephalography (EEG) is widely used in clinical trials and scientific research to measure the brain activities. However, analyzing and modeling scalp EEG signals is still an open field due to the complex and non-stationary nature of the EEG signal itself as well as the transformed signals. Therefore, in this paper, we propose a topological-based processing pipeline that utilizes persistent homology to capture the underlying system dynamic of the transformed EEG signals and further construct machine learning classifiers. A public available scalp EEG data is used to validate our algorithms, and the results show that the topological features successfully capture the subtle changes in the time-frequency representations revealed by Hilbert-Huang Transformation, with area under ROC curve reaching 0.96.