Pith. sign in

REVIEW 4 major objections 6 minor 16 references

Normative Cerebral Perfusion Across the Lifespan

T0 review · 4 major / 6 minor · reviewed 2026-08-08 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read Brain blood flow peaks around age 7 and then declines for life, and deviations from this curve can flag disease-specific perfusion abnormalities.

desk verdict A potentially citeable normative CBF resource that deserves a serious referee; the peak-age estimate rests on an untested sequence-by-age additivity assumption that the Discussion itself concedes. read the letter →

arxiv 2502.08070 v1 pith:VRNEMUOP submitted 2025-02-12 q-bio.NC

Xinglin Zeng , Yiran Li , Lin Hua , Ruoxi Lu , Lucas Lemos Franco , Peter Kochunov , Shuo Chen , John A Detre
show 1 more author
Ze Wang
This is my paper · ORCID
classification q-bio.NC
keywords cerebralbloodflowarterialspinlabelingMRInormativemodelinglifespanbraintrajectoriesGAMLSSAlzheimer'sdiseasemildcognitiveimpairmentneurodevelopment
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper sets out to give cerebral perfusion—the rate at which blood delivers oxygen and nutrients to brain tissue—a normative chart covering the entire human lifespan, from birth to 85 years. Using arterial spin labeling (ASL) MRI scans from 12,633 participants, of whom 9,363 passed quality control, the authors model cerebral blood flow as a smooth function of age. The resulting trajectories show a rapid postnatal rise, a global peak at approximately 7.1 years (95% CI: 6.5–7.5), and a gradual near-linear decline into old age, with regional peaks ranging from 6.2 to 8.6 years. The paper then converts each person's position on these curves into a deviation score and uses those scores to detect disease-specific perfusion abnormalities in Alzheimer's disease, frontotemporal dementia, mild cognitive impairment, and major depressive disorder. If the model holds, perfusion becomes a quantitative, age- and sex-adjusted biomarker that could flag abnormal brain development or neurodegeneration from a single scan.

What carries the argument

The machinery is distributional regression using a generalized additive model for location, scale, and shape, applied with a four-parameter Generalized Beta 2 distribution. The median and the coefficient of variation are each modeled as smooth B-spline functions of age, with sex and ASL sequence type as fixed effects and dataset site as a random effect, so the model captures both the mean growth curve and the way between-person variability changes across life. Individual deviation scores are computed by converting a person's fitted GB2 quantile into a standard normal z-score; extreme negative deviations (z < −2.3) are counted regionally and used to define MCI subtypes, classify disorders, and model longitudinal progression. Peak ages come from the first derivative of the modeled median curve crossing zero, and all curves are validated with bootstrap, split-half, balanced-resampling, and leave-one-site-out analyses.

What would settle it

One could re-fit the GAMLSS models while allowing age-by-sequence-type interaction terms and recompute the global peak; if the peak moves outside the 6.5–7.5 year interval, the 7.1-year peak is an artifact of pooling. A complementary check is to estimate the same curves using only sites with a single uniform ASL sequence and compare the resulting peak age and slope.

Watch

Extended reading notes

Core claim

The central discovery is a normative, age- and sex-aware lifespan trajectory for human cerebral blood flow: global CBF rises sharply after birth, peaks at about 7.1 years (95% CI: 6.5–7.5), and then declines gradually and near-linearly through adulthood. Cortical perfusion peaks highest in childhood; white matter peaks earlier (about 5.5 years) and declines more slowly; subcortical perfusion peaks around 6.7 years. Females show significantly higher perfusion than males in global, cortical, white matter, and subcortical measures, with a significant variance difference only in white matter. Individual deviation z-scores derived from the model reveal disease-specific hypoperfusion patterns, and classification analyses separate frontotemporal dementia (mean AUC 87.67) and Alzheimer's disease (mean AUC 80.24) from controls. In mild cognitive impairment, an AD-like perfusion subtype shows faster cognitive decline and higher odds of conversion to AD than a normal-perfusion subtype.

Load-bearing premise

The model assumes that pooling cross-sectional ASL scans from 20 sites with different acquisition sequences, and then adjusting for sequence type as a fixed effect and site as a random effect, yields unbiased age trajectories; if site or sequence effects interact with age, the estimated peak age of 7.1 years, the decline shape, and all patient deviation scores would be biased.

Editorial extensions

If this is right

  • If the model is correct, a single ASL scan can be compared against an age- and sex-matched reference, allowing a child with abnormally low perfusion or an adult with accelerated decline to be identified against a quantitative standard.
  • Perfusion milestones can be aligned with structural milestones, such as global perfusion peaking at 7.1 years and gray matter volume peaking at 7.8 years, giving developmental and aging research a shared temporal benchmark.
  • Perfusion deviation scores provide a transdiagnostic measure: AD and FTD separate from healthy controls with mean AUCs of 80.24 and 87.67, while MCI divides into AD-like and normal-perfusion subtypes with different conversion rates.
  • Longitudinal modeling shows that the proportion of extreme negative perfusion deviations increases faster in AD than in controls and faster in progressive than in stable MCI, supporting perfusion-based monitoring of disease progression.
  • Because sequence type and site are explicit terms in the model, the normative chart can in principle be reused across heterogeneous multi-site datasets, although the paper acknowledges sequence-related variation as a limitation.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Editorial inference: the pooling design can be stress-tested by adding age-by-sequence interaction terms; if the global peak then moves outside the reported 6.5–7.5 year interval, the 7.1-year peak would be a harmonization artifact rather than a biological milestone.
  • Editorial inference: the regional maturation gradient implies a testable prediction outside the paper—children with later prefrontal perfusion peaks should show later maturation of executive functions.
  • Editorial inference: the high between-subject perfusion variability seen in childhood suggests that deviation scores early in life may be less reliable than adult scores, so clinical use in children would require age-specific reliability estimates.
  • Editorial inference: the same normative machinery could be extended to other ASL-derived physiological measures, such as arterial transit time or blood-brain-barrier water exchange, which the paper notes were not available in most of its datasets.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 6 minor

Summary. The paper assembles a large multi-site ASL dataset and constructs normative models of cerebral perfusion across ages 0-85 using GAMLSS with a GB2 distribution. After quality control, the sample includes 8,460 healthy controls and 903 patients from 20 datasets. The central results are a global CBF peak at approximately 7.1 years (95% CI 6.5-7.5) followed by a near-linear adult decline, region-specific peak ages, sex differences, and normative deviation scores applied to AD, FTD, MCI, and MDD. Sensitivity analyses include stricter QEI thresholds, balanced resampling, split-half replication, bootstrap, and leave-one-site-out. The paper also reports longitudinal progression trajectories in AD and MCI, including MCI-A and MCI-N subtypes.

Significance. If the normative trajectory is unbiased, this would be a valuable resource analogous to structural brain charts, providing age- and sex-adjusted CBF norms and individualized deviation scores. The scale of the healthy-control sample, the explicit external-split structure for clinical applications, and the battery of sensitivity analyses are genuine strengths. The longitudinal MCI subtype analysis uses external outcomes (conversion, cognitive decline) and is a useful demonstration of prognostic value. However, the central age trajectory rests on an untested additivity assumption for site and ASL sequence effects, and the claimed birth-to-85 coverage is stronger than the acknowledged sparse sampling supports. The clinical significance is therefore conditional on the resolution of these methodological points.

major comments (4)
  1. [Methods, Eqs. (3)-(4); Discussion, first limitation] The model identifies the age trajectory only if site and ASL sequence effects are age-invariant. The fixed sequence term and the random site intercept in Eqs. (3)-(4) shift the mean across all ages equally and cannot absorb sequence-by-age or site-by-age interactions. Because pediatric and adult cohorts differ systematically in ASL sequence type, PLD, and M0 handling, and because arterial transit time changes with age, such interactions would bias the shape of the curve and the estimated peak age. The LOSO and balanced-resampling analyses do not test the additivity assumption itself. Please provide a direct test, for example by including an age-by-sequence interaction or site-specific age terms and reporting how the peak age and decline rate change, or by presenting within-site age trajectories for sites with wide age coverage.
  2. [Discussion, fourth limitation; Methods, balanced resampling] The claimed lifespan range 'birth to 85 years' overstates the empirical support because ages 0-5 and 30-40 are acknowledged to be underrepresented. Balanced resampling to 64 participants per five-year bin does not add information in sparse bins; it only equalizes weights. Please report the number of healthy participants in each five-year age bin (especially 0-5) in the main text or a main-table version of Supplementary Table 2, and rerun the global model excluding the sparse bins to quantify how dependent the 7.1-year peak and the adult decline are on those observations.
  3. [Results, Figure 3E; Methods, MCI subtype identification] The baseline TNP difference between the MCI-A and MCI-N subtypes is partly expected by construction, because the subtypes are obtained by k-means clustering on the same regional and network deviation scores used to compute TNP. The statement that 'Baseline TNP was significantly higher in MCI-A compared to MCI-N' is therefore a clustering check rather than an independent clinical finding. The independent evidence is the difference in longitudinal cognitive decline and conversion risk, which should be emphasized as the primary support for subtype validity; the baseline TNP contrast should be reframed accordingly.
  4. [Results, Figure 3F; Abstract] The abstract states that the model identifies 'disease-specific perfusion abnormalities across four brain disorders,' but the MDD classification performance is not significant (mean AUC = 54.88, p = 0.272). The post hoc explanation of imbalanced sample size is plausible but is not demonstrated. Please either soften the abstract and general claims to three disorders with significant classification, or provide a matched-control analysis showing that the MDD null result is attributable to sample imbalance.
minor comments (6)
  1. [Abstract vs. Results] The abstract says 'over 12,000 high-quality ASL MRI scans,' while the Results report 9,363 participants after QC. Clarify whether these are scans or participants, and reconcile the numbers.
  2. [Methods, Eq. (2)] Equation (2) uses the symbol gamma for the response variable, which is inconsistent with the text describing CBF values as y. This appears to be a typographical issue.
  3. [Introduction] The sentence 'Because ASL MRI does need any exogenous contrast agents' should read 'does not need.'
  4. [Methods, atlas regions; Methods, longitudinal TNP] The atlas description states 48 cortical and 8 subcortical regions, while the longitudinal TNP definition uses '58 cortical and 8 subcortical regions.' Since TNP is a proportion, the denominator must be consistent; please correct this discrepancy.
  5. [Methods, longitudinal dataset description] The sentence 'Prevent AD and Dalas lifespan were set as reference health control dataset... the ADNI were set as disease dataset, included 649 health controls with 213 scans' contains likely typos ('Dalas' and the numbers 649/213), and the dataset names should be spelled consistently.
  6. [Figure 1B and Methods, Eq. (3)] Figure 1B is labeled as sex-specific trajectories, but Eq. (3) includes sex only as an additive fixed effect, which would produce parallel curves. If sex-by-age interactions were not modeled, the figure legend should clarify that sex differences are constant offsets.

Circularity Check

1 steps flagged · score 4.0 of 10

The lifespan normative CBF model is externally structured, but the MCI-A vs MCI-N baseline TNP comparison is circular because TNP summarizes the same regional deviation scores used to create the clusters.

  1. fitted input called prediction [Results, 'Longitudinal Cerebral Perfusion in brain disorders'; Methods, '(ii). Identification of MCI Subtypes Using individual perfusion deviations' and '(i). Deviation z scores at each scan']
    "Baseline TNP was significantly higher in MCI-A compared to MCI-N (Estimate = 6.41, t = 4.527, p < 0.001). ... Deviation features for each patient included both regional and network-level CBF metrics. The Euclidean distance was used to calculate the similarity matrix across patients. The TNP was defined as the proportion of extreme negative deviations (z < −2.3) across all brain regions."

    TNP is a summary of the same regional deviation z-scores that were the input features to the k-means clustering. The clusters were then labeled MCI-A and MCI-N based on whether their deviation patterns resembled AD or CN. Consequently, the reported baseline TNP difference between MCI-A and MCI-N is not an independent biological finding: it recapitulates the clustering input by construction. Comparing clusters on a function of the exact variables used to define the clusters cannot provide confirmatory evidence about those clusters.

full rationale

The core normative model is built by fitting GAMLSS to healthy-control ASL CBF (Eqs. 2-6) and then applying the fitted model to held-out controls and patients, with the train/test split repeated 100 times. The patient deviation scores, classification AUCs, and conversion predictions therefore do not reduce to the model inputs. The lifespan peak-age trajectory is a descriptive summary of the fitted curve rather than an out-of-sample prediction, so it is not circular in itself. The main circularity is confined to the MCI subtype analysis: k-means clusters are computed from regional and network deviation z-scores, and the same z-scores are then summarized into TNP and compared between MCI-A and MCI-N at baseline. That difference is forced by the clustering input. Because this is a secondary clinical illustration rather than the load-bearing normative trajectory claim, the overall circularity score is moderate rather than high.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

The central contribution is a data-driven normative description, not a mechanistic derivation. The free parameters are all GAMLSS fit parameters, none of which are presented as derived constants. The main assumptions concern cross-site harmonization and representativeness of healthy controls, both acknowledged in the Discussion. No new physical entities, forces, or conserved quantities are introduced.

free parameters (4)
  • Age B-spline coefficients for mu and sigma (df=8) = Not reported in preprint
    Define the shape of normative CBF curves and the derived peak ages (e.g., 7.1 years).
  • Sex fixed-effect coefficients (beta_mu, beta_sigma) = Not reported in text (Supplementary Tables 3-4)
    Drive the reported female > male CBF difference.
  • ASL sequence fixed-effect coefficients = Not reported
    Used to harmonize PASL vs pCASL and other sequence variations across 20 sites.
  • GB2 distribution parameters nu and tau (intercepts) = Not reported
    Set skewness and kurtosis of the fitted CBF distribution; chosen by BIC among 28 distributions.
assumptions (5)
  • domain assumption Cross-sectional pooling of multiple ASL datasets, with sequence as fixed effect and site as random effect, yields unbiased lifespan CBF trajectories.
    Used in Eqs. (3)-(4) for the GAMLSS model; if site or sequence effects interact with age, peak age and deviations would be biased.
  • domain assumption The healthy control sample is representative of the general population at each age.
    Normative charts and z-scores presume the training healthy controls define 'normal' CBF.
  • domain assumption GB2 distribution and B-spline df=8 adequately capture CBF distribution and age shape.
    Selected by BIC; if misspecified, quantile-based z-scores could be biased.
  • domain assumption General kinetic model (Eq. 1) converts ASL signal to physiological CBF units.
    Relies on assumed labeling efficiency, blood T1, partition coefficient, etc.; different implementations across sites.
  • domain assumption ASL-derived CBF is comparable across 3T scanners after fixed-effect adjustment.
    Underlies pooling 20 datasets; acknowledged limitation in Discussion.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Normative Cerebral Perfusion Across the Lifespan." pith.science (2026). https://pith.science/paper/VRNEMUOP

@misc{pith2026250208070,
  author       = {Pith},
  title        = {Pith review of: Normative Cerebral Perfusion Across the Lifespan},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/VRNEMUOP}},
  note         = {Machine review of arXiv:2502.08070}
}
read the original abstract

Cerebral perfusion plays a crucial role in maintaining brain function and is tightly coupled with neuronal activity. While previous studies have examined cerebral perfusion trajectories across development and aging, precise characterization of its lifespan dynamics has been limited by small sample sizes and methodological inconsistencies. In this study, we construct the first comprehensive normative model of cerebral perfusion across the human lifespan (birth to 85 years) using a large multi-site dataset of over 12,000 high-quality arterial spin labeling (ASL) MRI scans. Leveraging generalized additive models for location, scale, and shape (GAMLSS), we mapped nonlinear growth trajectories of cerebral perfusion at global, network, and regional levels. We observed a rapid postnatal increase in cerebral perfusion, peaking at approximately 7.1 years, followed by a gradual decline into adulthood. Sex differences were evident, with distinct regional maturation patterns rather than uniform differences across all brain regions. Beyond normative modeling, we quantified individual deviations from expected CBF patterns in neurodegenerative and psychiatric conditions, identifying disease-specific perfusion abnormalities across four brain disorders. Using longitudinal data, we established typical and atypical cerebral perfusion trajectories, highlighting the prognostic value of perfusion-based biomarkers for detecting disease progression. Our findings provide a robust normative framework for cerebral perfusion, facilitating precise characterization of brain health across the lifespan and enhancing the early identification of neurovascular dysfunction in clinical populations.

Figures

Figures reproduced from arXiv: 2502.08070 by the authors.

Figure 2
Figure 2. ) [PITH_FULL_IMAGE:figures/full_fig_p009_2.png] view at source ↗
Figure 3
Figure 3. Clinical Relevance of Perfusion Deviation Patterns in Brain Disorders. A) Violin plots illustrating deviation scores across global and network levels for AD, [PITH_FULL_IMAGE:figures/full_fig_p012_3.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

16 extracted references · 13 canonical work pages

  1. [1]

    Data Quality Validation: Analyses were repeated with a stricter quality control threshold (QEI > 0.15) to assess the impact of data quality

    work page

  2. [2]

    Participants and site numbers were matched uniformly by resampling 1,000 times

    Balanced Sampling: A balanced sampling strategy was employed to address potential biases from uneven sample and site distributions across ages. Participants and site numbers were matched uniformly by resampling 1,000 times

    work page

  3. [3]

    Reproducibility Testing : A split -half approach was used to validate the reproducibility of the results by dividing the dataset into two equal halves and comparing the outcomes

    work page

  4. [4]

    Bootstrap Resampling: A bootstrap resampling analysis (1,000 iterations) was conducted to examine the influence of data sample variability

    work page

  5. [5]

    17 The results of these validation strategies were quantitatively compared to the primary findings and demonstrated consistent growth patterns across all analyses

    Site-Specific Effects: A leave-one-site-out (LOSO) analysis was performed to evaluate the potential impact of specific sites on the observed patterns. 17 The results of these validation strategies were quantitatively compared to the primary findings and demonstrated consistent growth patterns across all analyses. Detailed results are provided in Supplemen...

    work page 2021

  6. [12]

    S., Dingna Duan, Stefanie Darnley, Ying Jing, Jian Zhang, Charles O'Brien, Anna Rose Childress

    https://doi.org/10.7554/eLife.78397 Ze Wang, J. S., Dingna Duan, Stefanie Darnley, Ying Jing, Jian Zhang, Charles O'Brien, Anna Rose Childress. (2017). A Hypo-Status in Drug Dependent Brain Revealed by Multi-modal MRI. Addiction Biology, 22(6), 1622-1631. http://www.ncbi.nlm.nih.gov/pubmed/25957794 Zhang, N., Gordon, M. L., Ma, Y., Chi, B., Gomar, J. J., ...

    work page arXiv 2017

  7. [228]

    https://doi.org/ https://doi.org/10.13104/imri.2022.26.4.220 Wang, Z. (2022). Arterial Spin Labeling Perfusion MRI Signal Processing Through Traditional Methods and Machine Learning. Investig Magn Reson Imaging , 26(4), 220 -228. https://doi.org/10.13104/imri.2022.26.4.220 Wang, Z., Aguirre, G. K., Rao, H., Wang, J., Fernández -Seara, M. A., Childress, A....

    work page doi:10.13104/imri.2022.26.4.220 2022

  8. [636]

    P.,…Lerman, C

    https://doi.org/https://doi.org/10.1016/j.nicl.2013.04.014 Wang, Z., Faith, M., Patterson, F., Tang, K., Kerrin, K., Wileyto, E. P.,…Lerman, C. (2007). Neural substrates of abstinence-induced cigarette cravings in chronic smokers. J Neurosci, 27(51), 41 14035-14040, PMC2153440. https://doi.org/27/51/14035 [pii] 10.1523/JNEUROSCI.2966-07.2007 Wang, Z., Fer...

    work page arXiv 2007

Show all 16 references
  1. [769]

    A., Weiying Dai, Maria A Fernandez-Seara, Jia Guo, Matthias Guenther, Jonas Schollenberger, Ananth J

    https://doi.org/10.1146/annurev.physiol.66.082602.092845 Luis Hernandez-Garcia, V. A., Weiying Dai, Maria A Fernandez-Seara, Jia Guo, Matthias Guenther, Jonas Schollenberger, Ananth J. Madhuranthakam, Henk Mutsaerts, Jan Petr, Qin Qin, Yuriko Suzuki, Manuel Taso, David L. Thom...

    2022

  2. [1209]

    J., Vannest, J., Lee, G., Hernandez-Garcia, L., Plante, E., Rajagopal, A., & Holland, S

    https://doi.org/10.1038/s41593-021-00904-7 Schmithorst, V. J., Vannest, J., Lee, G., Hernandez-Garcia, L., Plante, E., Rajagopal, A., & Holland, S. K. (2015). Evidence that neurovascular coupling underlying the BOLD effect increases with age during childhood. Hum Brain Mapp, 3...

    2015 doi

  3. [1734]

    D., Shinohara, R

    https://doi.org/10.1038/s41596-022-00696-5 Satterthwaite, T. D., Shinohara, R. T., Wolf, D. H., Hopson, R. D., Elliott, M. A., Vandekar, S. N.,…Gur, R. E. (2014). Impact of puberty on the evolution of cerebral perfusion during adolescence. Proceedings of the National Academy o...

    2014 doi

  4. [2012]

    Fourth, the age distribution of participants in the datasets was uneven, with certain age ranges, such as 0–5 years and 30–40 years, being underrepresented

    and others (Alsop et al., 2014) , high-resolution ASL CBF maps may become more available, allowing us to explore finer-grained cerebral perfusion atlases. Fourth, the age distribution of participants in the datasets was uneven, with certain age ranges, such as 0–5 years and 30...

    2021

  5. [2013]

    , far surpassing the adult metabolic demands. This elevated perfusion ensures the delivery of oxygen and nutrients necessary for dynamic plasticity underlying sensory, cognitive, and motor development (Hijman et al., 2024; Ouyang et al., 2024) . At the 18 network level, the ma...

    2023

  6. [5945]

    https://doi.org/10.1109/jbhi.2023.3312662 Liao, W., Wang, Z., Zhang, X., Shu, H., Wang, Z., Liu, D., & Zhang, Z. (2017). Cerebral blood flow changes in remitted early-and late-onset depression patients. Oncotarget, 8(44), 76214. Liu, W., Lou, X., & Ma, L. (2016). Use of 3D pse...

    2017

  7. [7012]

    A., Thomas, D

    https://doi.org/https://doi.org/10.1002/alz.14174 Vidorreta, M., Balteau, E., Wang, Z., De Vita, E., Pastor, M. A., Thomas, D. L.,…Fernandez-Seara, M. A. (2014). Evaluation of segmented 3D acquisition schemes for whole -brain high - resolution arterial spin labeling at 3 T. NM...

    2014 doi

  8. [7802]

    A., Hyland, J

    https://doi.org/10.1038/s41598-024-57660-4 Ouyang, M., Detre, J. A., Hyland, J. L., Sindabizera, K. L., Kuschner, E. S., Edgar, J. C.,…Huang, H. (2024). Spatiotemporal cerebral blood flow dynamics underlies emergence of the limbic- sensorimotor-association cortical gradient in...

    2024

Pith tools

Reviewed August 8, 2026 · model on record in the stance chip above.