Open reviews
July 24, 2022 — Open review of preprint by Amalric and Cantlon:
Amalric, M., & Cantlon, J. F. (2020). Entropy, complexity, and maturity in children’s neural responses during naturalistic mathematics learning. bioRxiv. DOI
Amalric and Cantlon measure the sample entropy of children’s fMRI response time series during two naturalistic educational films. They show that entropy tracks with “functional maturity” as indexed by the similarity (i.e. intersubject correlation) of children’s response time series to adults. This result is localized to brain areas involved in processing the content of the different videos (math and grammar). This manuscript is concise and well-written; my main comments have to do with making sure readers are confident in the entropy results. Note that the version of the manuscript I reviewed for a journal is slightly different from the current version on bioRxiv.
Major comments:
Any way the authors can help build an intuition as to what increased neural entropy may reflect (whether based on the current data or the literature) would be appreciated. Figure 1B and the discussion of the neuroscience literature using neural entropy in the Introduction are very helpful. I have a couple lingering questions. Do we know how closely entropy tracks naturalistic stimulus processing in adults? For example, does it track with boredom or mind-wandering? Does higher sample entropy in adults (or children) correlate with any behavioral metric of stimulus comprehension? (This could be an avenue for future work if not available in the current sample.) In terms of building an intuition for the neural entropy metric, is it worth plotting some response times series for low-entropy children, high-entropy children, and adults for a meaningful ROI? Are there any visible differences between these time series?
Following on the previous comment, I’m curious whether this neural entropy metric would correlate with confound variables like head motion; you could compute the framewise displacement (averaged across time) per subject and correlate it with neural entropy. This might complicate things because I suspect ISC is also partly correlated with head motion, but partialling out subject-level head motion might help convince readers the neural entropy is really “neural” (especially in kids, where head motion tends to be a problem).
The “functionally immature” regions identified by Kersey et al. (2019) seem to largely correspond to regions with high (or reliable) ISC in adults during movie-viewing. Does this undermine the interpretation at all? For example, is it the case that functionally immature regions are simply those where adults have reliable enough ISC to detect a difference from children? Do these functionally immature regions identified by ISC correspond to any other neural metrics of maturity?
I recognize that developmental samples are difficult to obtain, but the sample size here is pretty small. As far as I understand, you had 24 children, but dropped 6, resulting in a sample size of 18—is this correct? If so, this sample size of N = 18 should be made very clear in the writing. Correlations computed across 18 samples (e.g. Figure 3) should probably be interpreted with caution, and the authors should be explicit about this. I’m also curious: what criterion was used to exclude children for “excessive head motion”?
I didn’t find the simulation in Figure 4 super compelling. I understand that the point is to demonstrate that you can have time series with different entropy but high ISC and time series with similar entropy and low ISC—that’s fine. But introducing a simple phase shift to yield low ISC doesn’t seem very realistic; this is not something we would expect to observe in actual naturalistic neuroimaging data (unless we had a stimulus onset/alignment bug). In practice, I suspect both metrics track with noise in some way, and I’d be curious to see how the entropy metric behaves if you parametrically increase noise from 0 (r = 1) until r ≈ 0
Do children with low entropy tend to have high ISC with each other, but lower ISC with high-entropy children? Or are “low-entropy” children all different from each other in a way that would yield low ISC? (Here I’m wondering about Finn’s “Anna Karenina” model; Finn et al., 2020.)
I don’t have a very good intuition for how SampEn is computed, but a pattern lengths of 2 and 3 (TRs?) seems very brief. Might be worth adding another sentence on why these parameters might be “optimal” according to Easson & McIntosh, 2019. Figure 1B also suggests that there’s some binning procedure to discretize the continuous fMRI time series; can you expand a bit more on this?
Minor comments:
Page 2, line 46: Earlier in this paragraph, you use the example of alternating and random time series of 1s and 0s where both have the same mean and variance but different entropy. However, your example structured time series of 1s and 0s at line 46 does not have the same number of 1s and 0s, and so will not have the same mean and variance as your previous example. Is it possible to use an example with intermediate entropy that still yields equal mean and variance to the prior examples?
Page 4, line 56: I would provide the reference for the “immature” regions here as well (i.e. Kersey et al., 2019).
Page 6, line 2: Remind readers that “functional maturity” here means children’s ISC with adults.
Page 6, line 19: “independently defined math- and language-related ROIs”—defined how/by who? Maybe (re-)include the reference here.
Page 6, line 43: “if was marginally” > “it was marginally
Figure 3. Include x- and y-axis labels in panels A and B scatterplots.
Page 11, line 27: “SanpEn” > “SampEn”
Page 12, line 13: “Fischer” > “Fisher”
Great that the data are publicly available!
References:
Easson, A. K., & McIntosh, A. R. (2019). BOLD signal variability and complexity in children and adolescents with and without autism spectrum disorder. Developmental Cognitive Neuroscience, 36, 100630. DOI
Finn, E. S., Glerean, E., Khojandi, A. Y., Nielson, D., Molfese, P. J., Handwerker, D. A., & Bandettini, P. A. (2020). Idiosynchrony: from shared responses to individual differences during naturalistic neuroimaging. NeuroImage, 215, 116828. DOI
Kersey, A. J., Wakim, K. M., Li, R., & Cantlon, J. F. (2019). Developing, mature, and unique functions of the child’s brain in reading and mathematics. Developmental Cognitive Neuroscience, 39, 100684. DOI