Reader Comments
Post a new comment on this article
Post Your Discussion Comment
Please follow our guidelines for comments and review our competing interests policy. Comments that do not conform to our guidelines will be promptly removed and the user account disabled. The following must be avoided:
- Remarks that could be interpreted as allegations of misconduct
- Unsupported assertions or statements
- Inflammatory or insulting language
Thank You!
Thank you for taking the time to flag this posting; we review flagged postings on a regular basis.
closePotential and caveats of complexity measures of rs-fMRI
Posted by maggie2190 on 03 Apr 2014 at 17:32 GMT
In this paper by Ze Wang et al., the authors explored Sample Entropy (SampEn) as a measure of the complexity of rs-fMRI in a large yet heterogeneous sample. The results are certainly interesting. Complexity measures including Approximate Entropy (ApEn), SampEn and MultiScale Entropy (MSE) have been mainly applied for electrophysiological signals (EEG, ECG), and just begun to be explored for rs-fMRI. The field is very young and vulnerable, lack of clear guidance of which complexity measure to use with what parameters (e.g. m and r). I would raise 2 caveats for both the authors and the audience to consider:
1. SampEn has limitations, e.g. random thermal noise yields the highest SampEn but does not represent the most complex process. As a result, the authors detected higher entropy in WM vs. GM. Recent papers (Yang et al HBM 2013; Smith et al BIB 2014) using MSE reported more consistent results with greater entropy in GM vs. WM and also decreasing entropy with aging which is one of the most reliable findings in electrophysiological studies. The authors mentioned the length of their rs-fMRI data was not long enough for MSE analysis. Therefore the results need to be interpreted with caution.
2. SampEn as well as other complexity measures are sensitive to the length, SNR of signals as well as parameters of m and r. Ideally, one would like to adjust the tolerance level r to be just above random noise level. To apply the same m, r to a heterogeneous rs-fMRI sample with different signal length and SNR (field strength) is likely to include large variations in measurements that swamp effects of aging and gender etc.
Overall, I feel the results of this paper should be interpreted with caution, and more advanced complexity measures such as MSE and wavelet based MSE should be applied for rs-fMRI acquired with adequate length and SNR (e.g. with the latest multiband EPI sequence).
Danny JJ Wang
UCLA Neurology
RE: Potential and caveats of complexity measures of rs-fMRI
redhatw replied to maggie2190 on 03 Apr 2014 at 21:07 GMT
Thanks for commenting. But I guess you missed a point here: as spelled in the title, we are talking about entropy here rather than complexity. Those two are somehow related but can be very different. Complexity decreases after it reaches the maximum when entropy still goes up. If you read our long-time explorations on this direction including the abstracts: 1. Wang, Z. (2012a) Characterizing Resting Brain Information using Voxel-based Brain Information Mapping (BIM), 2012 Annual Meeting of the Organization for Human Brain Mapping. Beijing, China. 2. Wang, Z. (2012b) Stable and Self-Organized Entropy in the Resting Brain. The Third Biennial Conference on Resting State Brain Connectivity. Magdeburg, Germany. p 208, 3. Ze Wang, A.M., Marcus Raichle, Anna Rose Childress, and John A Detre. (2013a) Mapping brain entropy using resting state fMRI. 2013 Annual Meeting of International Society of Magnetic Resonance in Medicine. Salt Lake City, USA. p 4861. 4. Ze Wang, J.S., Y. Li, Z. Singer, R. Ehrman, A. V. Hole, C. P. O'Brien, Anna Rose Childress. (2013b) Human brain entropy mapping using thousands of subjects and its application in a drug addiction study. 2013 Annual Meeting of Society for Neuroscience. San Diego. p 7491, we were all talking about entropy. Since complexity increases with entropy before it reaches the peak, it is possible that entropy may be used for complexity assessment if we know the peak, but that is not the focus of this paper.
Regarding to your point of higher entropy in grey matter but lower in white matter as found in your group's paper, if you thought is as "entropy", I would argue the opposite way because it is well known that white matter signal is more random and the higher randomness means higher entropy. And that is exactly what we showed here. We didn't use MSE because the data length is short also because we wanted to focus on entropy rather than getting more complicated. But not doing MSE should never mean that the non-MSE results are not reliable. If you see the appendix, you can see that the mean entropy pattern stay across different m and r.
For the 2nd question, our evaluations with simulations, task-fMRI, and test-retest data should be enough to answer it. With the comprehensive evaluations and the 1049 sample size, we are confident to say that the results are reliable and insensitive to the parameter selections.
We have processed the freely available multi-band data. But haven't wrapped it to a paper. If anyone is interested in trying that, you are free to do it. We can provide our C++ based compiled code for measuring many different types of entropy including sample entropy (including multi-scale), approximate entropy, LZ... etc. It can compute entropy for a whole brain within 10 sec. Our matlab code would take hours but we can share as well.
In summary, we thank for bringing up the concept of complexity here though our paper was about entropy. We are fully confident on our results based on the comprehensive evaluations and large sample size. We are happy to share our software.