[NBThelp] DFA exponent in fNIRS signals

[<hui...@fo...>]

Hello everyone,
      I'm a PhD candidate in the School of Biological Sciences & Medical Engineering, Southeast University, Naning, China. My research interest is mainly focused on scale-free analysis of neural signals collected from young children with autism. 
   
      One of my research project is to investigate the scale-free features of fNIRS signals in children with autism. In the preprocessing of fNIRS signals, a commonly used procedure is performing a band-pass filtering (0.01 ~ 0.1 Hz), in order to suppress the typical noise components in fNIRS signals. It's known that filtering using FIR, would introduce correlation in the signal between the neighboring samples. Thus, if a 0.01 Hz high-pass filtering is conducted, the window sizes in DFA exponent estimation will be too large (i.e., at least 300 sec, when the filter order for the FIR filter is set to three cycles of 0.01 Hz). However, if this high-pass filtering is not performed, the fNIRS signals will contain the low-frequency drift, which is very common in fNIRS signals. The DFA exponent estimated may be distorted due to this noise component.


     In the data analysis, only a 0.1 Hz low-pass filtering is conducted. The low-frequency drift has been partly removed through ICA, although the ICA could not fully remove this artifact. This DFA exponents calculated in our study were larger than 1, mainly distributed among 1 and 1.5. This result was very weird, since as far as I know, for many kinds of physiological signals, the DFA exponent is larger than 0.5 and smaller than 1. Does this result suggest that the fNIRS signals are highly non-stationary?


    Could you give me some advice about the two questions above?
    Thank you!




    Huibin Jia
PhD candidate
School of Biological Sciences & Medical Engineering, Southeast University, Naning, China