Exploring and Organizing Spatiotemporal Features such as Waves in High Throughput Brain Recordings by Lifting to Feature Space

High throughput techniques for recording brain signals and other medical processes produce data that is difficult to analyze because of noise, complexity and massive volume. If this data has correlations in short time segments, the segments can be represented by low-dimensional features and the dataset can be organized by feature characteristics. We explore the possibility of lifting this type of data to a feature space and develop techniques for exploration of that space. We apply the techniques to data characterized by short-duration waves traveling in many directions both from synthetically generated datasets and from multielectrode brain recordings. Projection and navigation techniques, which summarize the distribution and relationships of features across large datasets, can be used in conjunction with color mapping and sorting strategies to compare feature geometry. The key is to find a feature space and a distance metric that emphasize interesting aspects of the data.