A reduced time-series feature library to efficiently characterize neural dynamics
Psychiatric disorders are diagnosed based on symptom scores from clinical interviews, there are no existing gold standards that can be used for definitive validation. Brain functional neuroimaging techniques including functional magnetic resonance imaging (fMRI), Positron Emission Tomography (PET), and Electroencephalography (EEG) have become important tools in investigating brain disease. Thus, the analysis of functional neuroimaging data can be used to characterize brain function abnormality.
Recently the researchers have formulated different ways to analyze time-series data. Some sophisticated and some simple. Although the simple methods work quite well, there is a need to apply the complex methodologies for analysis. In the paper published by Fulcher et. al, 2017 ( hctsa: A Computational Framework for Automated Time-Series Phenotyping Using Massive Feature Extraction ), showed that numerous features (~7700) can be extracted from a time-series using the proposed hctsa tool. But using the hctsa features for analysis can be computationally expensive and requires closed Matlab licenses to run, limiting widespread adoption for medical and research applications. These features can be reduced to a lesser number of non-redundant features to represent the time-series, using the catch22 feature reduction approach ( catch22: CAnonical Time-series CHaracteristics ). The reduction is performed with minimal loss of the classification accuracy. In this method, the performance of each feature was evaluated on a set of datasets that are very different to NeuroImaging data. Thus, the current work will validate the proposed approach with NeuroImaging data.
This project is particularly important for measurements of brain dynamics of a patient to their disease diagnosis. It further reduces complexities by distilling a large literature on time series analysis into a small subset. It selects only the significant features to represent the time series for analysis, with minimal loss in classification accuracy. The optimized and efficiently coded features will be made into a package for the community to use it. This open source code will let other researchers find the underlying hidden features of NeuroImaging time series and utilize it in their work. The availability and ease of computing the features opens up the application to several areas.
- Reduce complexities by distilling a large literature on time series analysis into a small subset
- Minimize loss of classification accuracy by using only significant features to represent the time series, with minimal loss in classification accuracy
- Package coded features for community use
- Allow researchers to find underlying hidden features of NeuroImaging time series