Abstract:
|
Understanding the development of intrinsic brain network organization based on resting state functional Magnetic Resonance Imaging (rs-fMRI) is of increasing importance and interest. However, given a sample of rs-fMRI images from different aged subjects, the problem of how to test for changes in network organization across ages is not well understood. We built two new approaches, eigen test (ET) and likelihood ratio test (LRT), to fill in the gap. As the names suggest, ET utilizes the eigenvectors and LRT compares the log likelihood of competing models. The former is sensitive only to changes in community assignment, while the latter detects both changes in community assignments and connectivity between communities. We are interested detecting various changes in network evolution: smooth trends over time, multiple change points and outliers. We also developed a theoretical framework for LRT, especially for models with multiple change points. We evaluated our methods in a wide variety of settings: simulated data, rs-fMRI data and dynamic gene co-expression networks, which can be analyzed with structure similar to rs-fMRI networks.
|