Masters candidate in Biostatistics, Trang Le, will present:

“Bayesian Spectral Estimation and Test-retest Reliability of Low-frequency Oscillation Amplitudes in Resting-state fMRI”

Plan B Adviser: Mark Fiecas

Abstract: The amplitude of low frequency oscillations in spontaneous fluctuations of brain signals obtained from resting-state functional magnetic resonance imaging (rs-fMRI), as quantified from power spectral density (PSD), is a measure of brain activity with the potential to detect brain abnormality. The present work provides a Bayesian model for using smoothing splines to estimate the PSD of rs-fMRI signals. Spectral power within frequency bands of interest are summary measures known as ALFF (amplitude of low frequency fluctuations) that can extracted from the estimated PSDs and used in a subsequent test-retest analysis. In this analysis, the three frequency bands of interest are the low-frequency ((0.01-0.08) Hertz), the slow-4 ((0.027-0.073) Hertz), and the slow-5 ((0.01-0.027) Hertz) frequency bands. After extracting the average power within each frequency band, I conducted a subject-level and a population-level test-retest reliability analysis to examine the stability of the ALFF across scan sessions. The results show that the slow-5 band had the most reliable PSD among the three subdivisions in both subject-level and population-level reliability. However, overall, ALFF yielded low test-retest reliability across scan sessions.