TY - JOUR
T1 - Filtered correlation and allowed frequency spectra in dynamic functional connectivity
AU - Vergara, Victor M.
AU - Calhoun, Vince D.
N1 - Funding Information:
This work was funded by the following NIH grants P20GM103472 / 1R01EB006841 / R01MH118695 / R01EB020407 and by the National Science Foundation (# 1539067 ) to V.C.
Publisher Copyright:
© 2020 Elsevier B.V.
PY - 2020/9/1
Y1 - 2020/9/1
N2 - Background: Dynamic functional connectivity enables us to study brain connectivity occurring at different frequencies. Techniques like sliding window correlation allow for the estimation of time varying connectivity and its frequency spectrum content. Since correlation is equal to the cosine of the phase (cos θ) between activation amplitudes of two brain regions, we assume that phase is the relevant functional connectivity feature and leave out any contamination from activation amplitudes. New method: First, this work studies the conditions by which time varying correlation can be separated from nuisance activation amplitudes that are not phase related. Second, we propose the filtered sliding window correlation to perform time varying estimation of cosine-phase (cos θ (t)) and nuisance filtering in one single step. Results: Mathematical models predict the correlation frequencies that should be filtered out to avoid overlap with the activation amplitude spectra. Filtered sliding window correlation excluded nuisance frequencies with an accurate estimation of time varying correlation. Real data outcomes empirically suggest that fMRI frequencies of interest extend up to 0.05 Hz. Comparison with existing methods: Compared with sliding window methods, the filtered sliding window correlation achieves better estimation for frequencies of interest. Conclusions: The filtered sliding window correlation approach allows controlling for nuisance frequencies unrelated to time varying phase estimation.
AB - Background: Dynamic functional connectivity enables us to study brain connectivity occurring at different frequencies. Techniques like sliding window correlation allow for the estimation of time varying connectivity and its frequency spectrum content. Since correlation is equal to the cosine of the phase (cos θ) between activation amplitudes of two brain regions, we assume that phase is the relevant functional connectivity feature and leave out any contamination from activation amplitudes. New method: First, this work studies the conditions by which time varying correlation can be separated from nuisance activation amplitudes that are not phase related. Second, we propose the filtered sliding window correlation to perform time varying estimation of cosine-phase (cos θ (t)) and nuisance filtering in one single step. Results: Mathematical models predict the correlation frequencies that should be filtered out to avoid overlap with the activation amplitude spectra. Filtered sliding window correlation excluded nuisance frequencies with an accurate estimation of time varying correlation. Real data outcomes empirically suggest that fMRI frequencies of interest extend up to 0.05 Hz. Comparison with existing methods: Compared with sliding window methods, the filtered sliding window correlation achieves better estimation for frequencies of interest. Conclusions: The filtered sliding window correlation approach allows controlling for nuisance frequencies unrelated to time varying phase estimation.
KW - Dynamic functional connectivity
KW - Filtering
KW - Sliding window correlation
UR - http://www.scopus.com/inward/record.url?scp=85087766202&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85087766202&partnerID=8YFLogxK
U2 - 10.1016/j.jneumeth.2020.108837
DO - 10.1016/j.jneumeth.2020.108837
M3 - Article
C2 - 32621916
AN - SCOPUS:85087766202
SN - 0165-0270
VL - 343
JO - Journal of Neuroscience Methods
JF - Journal of Neuroscience Methods
M1 - 108837
ER -