Pith. sign in

REVIEW 4 major objections 4 minor 2 cited by

Classification of Mild Cognitive Impairment Based on Dynamic Functional Connectivity Using Spatio-Temporal Transformer

T0 review · 4 major / 4 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read A spatio-temporal transformer that jointly encodes the timing and spatial layout of dynamic brain connectivity, boosted by contrastive learning, reaches 89.1% accuracy in distinguishing mild cognitive impairment from normal controls on…

desk verdict A sound internal story—subject-level CV and clean ablations—but the central 'superiority' claim rests on cross-study comparisons that don't match on data, atlas, or sample size. read the letter →

arxiv 2501.16409 v1 pith:WL4AISUW submitted 2025-01-27 eess.IV cs.AIq-bio.NC

Jing Zhang , Yanjun Lyu , Xiaowei Yu , Lu Zhang , Chao Cao , Tong Chen , Minheng Chen , Yan Zhuang
show 2 more authors
Tianming Liu Dajiang Zhu
This is my paper · ORCID
classification eess.IVcs.AIq-bio.NC
keywords mildcognitiveimpairmentdynamicfunctionalconnectivityresting-statefMRIspatio-temporaltransformercontrastivelearningAlzheimer'sdiseasebrainstateclassificationADNI
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 show that mild cognitive impairment (MCI), the prodromal stage of Alzheimer's disease, can be distinguished from healthy aging by modeling how brain connectivity changes over time, rather than by a single static connectivity map. The authors build a transformer-based network that reads the sequence of sliding-window functional connectivity matrices twice, once as a temporal trajectory and once as a spatial layout of regions, then fuses the two views with global attention. A contrastive learning step, in which scans from the same diagnosis are pulled together and scans from different diagnoses are pushed apart, is added to reduce dependence on labeled data. On 345 subjects and 570 scans from ADNI, the full model achieves 89.1% accuracy, 91.4% sensitivity, 87.2% specificity, and 89.3% AUC. The paper claims these results show the framework outperforms earlier static and dynamic connectivity classifiers and has clinical potential for early AD identification.

What carries the argument

The load-bearing object is the sequence of dynamic functional connectivity matrices $M = \{M_1,\dots,M_T\}$ built by sliding a window of length $L=70$ with step $S=2$ over each subject's ROI-averaged time series, with each $M_t$ holding Pearson correlations among 148 Destrieux regions. The machinery is a spatio-temporal transformer block applied twice: a temporal block reads each matrix transposed, so rows are time points, while a spatial block reads rows as regions. Each block projects its input to query, key, and value matrices, computes multi-head self-attention with $\mathrm{softmax}(QK^\top/\sqrt{d})V$, applies a feed-forward network with residual connections and layer normalization, then a convolutional layer. The two views are concatenated and fused by global attention before a single-layer perceptron classifies the subject. The contrastive branch defines positive pairs as dFC networks from subjects with the same diagnosis and negative pairs as those from different diagnoses, optimizing an InfoNCE-style loss with temperature $\tau$; the total loss is $\alpha\mathcal{L}_{\mathrm{cont}} + \beta\mathcal{L}_{\mathrm{CE}}$ with learnable weights.

What would settle it

Re-run one of the Table 1 baselines, for example the deep spatial-temporal fusion method [24], on the exact same 345 subjects, same Destrieux atlas, same sliding-window parameters ($L=70$, $S=2$), and same subject-level 5-fold split. If that baseline's accuracy is statistically indistinguishable from or above the reported 89.1%, the claim that the proposed architecture is superior would be falsified; if it remains clearly lower, the claim would gain direct support.

Watch

Extended reading notes

Core claim

The central claim is that the temporal sequence of connectivity matrices contains information that static functional connectivity discards, and that a transformer can jointly capture both the temporal and the spatial structure of that sequence. The architecture processes each windowed connectivity matrix in two orientations: a temporal block sees the time course of each region, and a spatial block sees the region-to-region coupling within each window. Both blocks combine a multi-head self-attention transformer layer with a convolutional layer, and their outputs are concatenated and passed through global attention to form a fused whole-brain representation. Training adds a contrastive loss, defined by cosine similarity between dFC representations, to the cross-entropy classification loss. The paper reports that this full model reaches 89.1% accuracy, and that the contrastive component improves accuracy by 9.7 percentage points over the same architecture trained without it.

Load-bearing premise

The claim that the proposed method is superior to previous work assumes that the previously published results in Table 1 are directly comparable to the new model's numbers, even though those studies used different numbers of subjects, different MCI subtypes, a different brain atlas, and different preprocessing, and no matched experiment is reported.

Editorial extensions

If this is right

  • If the central claim holds, MCI screening can be performed directly from the sequence of resting-state connectivity, giving 89.1% accuracy and 91.4% sensitivity on the ADNI cohort and catching most MCI cases at the prodromal stage.
  • The contrastive branch, which pairs scans by diagnosis and pulls same-diagnosis representations together, is reported to add 9.7 percentage points of accuracy over the same architecture without it, implying that diagnostic pairing provides a training signal beyond the label itself.
  • The ablation study implies the temporal and spatial branches are complementary: spatial-only reaches 76.6% accuracy, temporal-only 70.7%, and the fused model 89.1%, so the joint encoding carries the performance.
  • At 87.2% specificity, roughly one in eight healthy controls would be flagged for further testing, a reasonable screening operating point if the numbers transfer to independent cohorts.

Reading between the lines

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

  • The paper leaves untested whether the same architecture transfers to the EMCI subgroup or to independent cohorts, since the Table 1 comparisons mix EMCI and MCI samples across different studies.
  • A natural next step, not reported here, is to pretrain the temporal and spatial transformers on unlabeled rs-fMRI scans with the contrastive objective and then fine-tune on labeled MCI data; the reported contrastive gain suggests a fully self-supervised stage could push accuracy further.
  • Because window length and step size were fixed at $L=70$, $S=2$, a reader interested in the mechanism would want a sweep of these parameters: if performance is highly sensitive to them, part of the reported superiority may reflect the window configuration rather than the architecture.
  • A matched head-to-head re-implementation of the Table 1 baselines on the same 345-subject, Destrieux-parcellated, subject-level 5-fold split would be the direct test of the superiority claim; the paper does not provide that experiment.
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 / 4 minor

Summary. The paper proposes a spatio-temporal transformer with a contrastive learning loss for classifying mild cognitive impairment (MCI) versus normal control (NC) from resting-state fMRI. Dynamic functional connectivity (dFC) networks are constructed with a sliding-window strategy over the Destrieux atlas time series, and a temporal block and a spatial block, each containing a transformer layer and a convolutional layer, are used to extract fused representations. A contrastive loss that pulls together dFC features from subjects with the same diagnosis and pushes apart features from different diagnoses is combined with a cross-entropy classification loss. The authors report subject-level 5-fold cross-validation on 345 ADNI subjects (570 scans) and compare with ten literature results in Table 1, claiming superiority with accuracy 89.1%, sensitivity 91.4%, specificity 87.2%, AUC 89.3%, and F1 90.3% for the full model with contrastive learning. Ablation experiments in Table 2 compare static FC, spatial-only, temporal-only, and the full model.

Significance. If the reported superiority held under a rigorous comparison protocol, this would be a practically useful contribution to early AD screening, and the general design of jointly modeling temporal and spatial dFC information in a transformer is interesting. The paper has clear strengths: the subject-level 5-fold cross-validation prevents scan-level data leakage; the ablation study is a reasonable internal check; and the contrastive-learning mechanism is concretely described with a joint loss. However, the central empirical claim of superiority over prior work is not yet supported by the evidence presented, because the comparisons in Table 1 are cross-study and uncontrolled. The contribution is therefore promising but requires substantial additional validation before the abstract-level claim is justified.

major comments (4)
  1. [Table 1 and Section 3.2] The central claim that the method "demonstrate[s] the superiority" on ADNI is not supported by the evidence. Each comparator row in Table 1 comes from a different publication with a different sample size (29 to 95 MCI/NC subjects versus 120 MCI/225 NC here), a different diagnostic subgroup (EMCI versus MCI), a different atlas (AAL versus Destrieux), and different preprocessing and window choices. None of these baselines is re-run on the same 345-subject/570-scan sample under the same subject-level 5-fold protocol, and no confidence intervals, standard deviations, or significance tests are reported. The 89.1% accuracy therefore reflects a cross-study comparison with uncontrolled confounds and cannot by itself establish superiority. The authors should run at least one or two strong baselines (e.g., a static-FC classifier, a standard dFC classifier, or the best-performing prior method) on the exact same data and protocol, and report matched results with statistical testing.
  2. [Abstract and Section 2.1.2] The claim that contrastive learning "reduc[es] dependency on labeled data" is misleading in its current form. Positive and negative dFC pairs are constructed explicitly from diagnosis labels ("same diagnosis status" versus "different diagnoses"), so the contrastive loss uses the same labels as the classification loss, just in a pairwise form. The paper does not demonstrate any ability to exploit unlabeled data. This claim should either be qualified to say that the contrastive loss provides additional supervisory signal from pairwise label structure, or the method should be extended and evaluated with truly unlabeled samples.
  3. [Table 2 and Section 3.2] The ablation study does not cleanly isolate the contribution of contrastive learning. Table 2 compares os-FC, S-only, T-only, and "Ours" (89.1%), but the "Ours" row includes the contrastive loss, so the improvement over S-only/T-only could be driven by contrastive learning rather than by the temporal-spatial architecture. Meanwhile, Section 3.2 attributes a 9.7% accuracy improvement to contrastive learning by comparing "Ours" without contrastive (82.4%, Table 1) with "Ours (cl)" (89.1%), but that no-contrastive full model is absent from Table 2. The ablation should include the full temporal-spatial model without contrastive learning, and the word "significantly" should be backed by a significance test rather than a single point comparison.
  4. [Section 2.1 and Section 3.1] The method is not sufficiently specified for reproduction. The paper gives window length L=70 and step size S=2, but does not report the number of windows T, the length of the ROI time series after preprocessing, the number of transformer layers, the number of attention heads, the embedding dimensions, the feed-forward hidden size, the pooling strategy before classification, or the specific values of the learnable hyperparameters alpha and beta in Eq. (8). These details are necessary for any independent verification of the reported 89.1% accuracy and for assessing whether the method is genuinely novel relative to existing dFC transformers.
minor comments (4)
  1. [Title and Abstract] The title contains a typo: "CLASSIFFICATION" should be "CLASSIFICATION," and the abstract uses "BLOD" where "BOLD" is intended.
  2. [Section 2] The subsection numbering is inconsistent: Sections 2.1.1 and 2.1.2 appear as subsubsections of Section 2.2 (Model), but they are numbered as if they belong to Section 2.1. They should be renumbered 2.2.1 and 2.2.2.
  3. [Equation (6)] The notation "1[NO#]" in Eq. (6) is unclear; the indicator function should be written explicitly, for example as 1[k ≠ i] with a definition of the pair index k, so that the contrastive loss is formally readable.
  4. [References] Reference [12] (Lin et al., Frontiers in Neuroscience, 2022) is listed twice, and reference [11] is missing from the reference list; the citation numbering should be corrected throughout.

Circularity Check

1 steps flagged · score 1.0 of 10

No load-bearing circularity; only a local overclaim that contrastive learning reduces label dependence, contradicted by its label-defined pairs.

  1. other [Sec. 2.1.2, Eqs. (5)-(6); Abstract; Conclusion]
    "where dFC generated from subjects with the same diagnosis status (e.g., both NC) form “same” pairs, designed to “attract” each other. Conversely, dFC networks from different subjects with different diagnoses (e.g., NC and MCI) form "different" pairs and "repel" each other. ... This loss is computed across all pairs, leveraging the inherent information within the data itself to generate supervisory signals without requiring additional manual label information."

    The claimed label-free contrastive supervision is defined by the diagnosis labels: Eq. (6)'s positive/negative indicator is exactly whether two dFCs come from subjects with the same diagnostic label, and the same labels feed the cross-entropy loss in Eq. (7). The 'inherent information' is therefore the manual label by construction, so the paper's stated benefit that contrastive learning reduces dependency on labeled data is a definitional overclaim. This does not force the reported accuracy, but the data-efficiency rationale is circular.

full rationale

This is an empirical supervised-learning paper with no equation-level derivation of the main result. The classification pipeline (dFC construction, transformer blocks, fusion, CE loss) is a standard architecture, and the ablation in Table 2 is a matched internal control; the reported ACC/AUC are not fitted predictions of the input labels in the sense of a derivation. The main external-support claim ('superiority' over prior methods) rests on Table 1, whose rows come from different samples, atlases, subgroups, and protocols; that is a comparability/validity problem, not circularity. No load-bearing self-citation chain or imported uniqueness theorem is present. The only circularity-adjacent passage is the claim that contrastive learning reduces dependence on labeled data: the positive/negative pairs are generated from the same diagnosis labels used for classification, so that specific benefit claim is contradicted by construction. Because this is a secondary interpretive claim rather than the source of the reported accuracy, the circularity score is low.

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

The paper introduces no new entities, forces, or conserved quantities. The central result depends on hand-set dFC parameters (L, S), an unreported temperature tau, learnable loss weights, and unspecified transformer hyperparameters. The domain assumptions are that the preprocessing and dFC construction preserve MCI signal, and that the baseline numbers in Table 1 are comparable to the present results, which is not established.

free parameters (5)
  • dFC window length L = 70
    Set empirically in Section 2.1; no sensitivity analysis over L is provided.
  • dFC sliding step S = 2
    Set empirically in Section 2.1; no sensitivity analysis over S is provided.
  • contrastive temperature tau = not specified
    Temperature in Eq. 6; its value is not reported in the paper.
  • loss weights alpha and beta = learnable
    Eq. 8 describes alpha and beta as learnable hyperparameters, but final trained values are not reported.
  • transformer architecture hyperparameters = not specified
    Number of heads, layers, embedding dimensions, and convolution kernel sizes are not stated, which are needed for replication.
assumptions (4)
  • domain assumption Sliding-window Pearson correlation captures MCI-relevant dynamic functional connectivity
    Section 2.1 constructs dFC matrices as PCC within each window; no validation that this representation preserves disease signal.
  • domain assumption ADNI preprocessing pipeline from [17,18] is correctly applied
    Section 2.1 delegates preprocessing to references; no details or quality checks are provided.
  • ad hoc to paper Different study results in Table 1 are directly comparable
    Table 1 mixes different sample sizes, atlases, diagnostic subgroups, and preprocessing; the paper treats them as comparable baselines.
  • standard math Softmax attention and transformer blocks are standard building blocks
    Equations 1 through 4 use standard multi-head self-attention and feed-forward networks without modification.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Classification of Mild Cognitive Impairment Based on Dynamic Functional Connectivity Using Spatio-Temporal Transformer." pith.science (2026). https://pith.science/paper/WL4AISUW

@misc{pith2026250116409,
  author       = {Pith},
  title        = {Pith review of: Classification of Mild Cognitive Impairment Based on Dynamic Functional Connectivity Using Spatio-Temporal Transformer},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/WL4AISUW}},
  note         = {Machine review of arXiv:2501.16409}
}
read the original abstract

Dynamic functional connectivity (dFC) using resting-state functional magnetic resonance imaging (rs-fMRI) is an advanced technique for capturing the dynamic changes of neural activities, and can be very useful in the studies of brain diseases such as Alzheimer's disease (AD). Yet, existing studies have not fully leveraged the sequential information embedded within dFC that can potentially provide valuable information when identifying brain conditions. In this paper, we propose a novel framework that jointly learns the embedding of both spatial and temporal information within dFC based on the transformer architecture. Specifically, we first construct dFC networks from rs-fMRI data through a sliding window strategy. Then, we simultaneously employ a temporal block and a spatial block to capture higher-order representations of dynamic spatio-temporal dependencies, via mapping them into an efficient fused feature representation. To further enhance the robustness of these feature representations by reducing the dependency on labeled data, we also introduce a contrastive learning strategy to manipulate different brain states. Experimental results on 345 subjects with 570 scans from the Alzheimer's Disease Neuroimaging Initiative (ADNI) demonstrate the superiority of our proposed method for MCI (Mild Cognitive Impairment, the prodromal stage of AD) prediction, highlighting its potential for early identification of AD.

Figures

Figures reproduced from arXiv: 2501.16409 by the authors.

Figure 1
Figure 1. Illustration of proposed Spatial-Temporal Transformer based on Contrastive Learning framework for brain disease classification with dFC, consisting of (A) imaging preprocessing and dynamic characterization of whole-brain connectivity construction. (B) Contrastive dFC pairs construction and feature extraction at Temporal and Region levels via Spatial-Temporal block. Both cross-entropy loss and contrastive loss are em… view at source ↗

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 2 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Bridging Brain Connectomes and Clinical Reports for Early Alzheimer's Disease Diagnosis

    cs.CV 2025-08 reject novelty 6.0 of 10

    A vision-language model that treats brain subnetworks as image tokens and aligns them with words in clinical reports reaches 92.94% MCI accuracy on ADNI, but label leakage and uncontrolled baselines undermine the claim.

  2. Domain-Adaptive Diagnosis of Lewy Body Disease with Transferability Aware Transformer

    cs.LG 2025-07 reject novelty 4.0 of 10

    TAT, a transferability-aware transformer, adapts an Alzheimer's model to Lewy Body Disease, but its LBD classification accuracy (14.5%) is below the 33% chance level.

Reference graph

Works this paper leans on

3 extracted references · 3 canonical work pages · cited by 2 Pith papers

  1. [2]

    ,…,head-)𝑊. (2) ℎ𝑒𝑎𝑑#(𝑄,𝐾,𝑉)=softmaxB*+!/0

    METHOD 2.1. Subjects and Image Preprocessing We used 345 subjects of the ADNI dataset [16], which included both the normal control (NC) group (88 males, 137 females; 73.21 ± 7.66 years) and 120 subjects from the MCI group (70 males, 50 females; 74.64 ± 8.32 years). Each subject’s resting-state fMRI (rs-fMRI) underwent the same standard pre-processing proc...

    work page

  2. [5]

    COMPLIANCE WITH ETHICAL STANDARDS This is a computational simulation study for which no ethical approval was required

    ACKNOWLEDGMENTS This work was supported by National Institutes of Health (R01AG075582 and RF1NS128534) 6. COMPLIANCE WITH ETHICAL STANDARDS This is a computational simulation study for which no ethical approval was required. 7. REFERENCES [1] Lin K, Jie B, Dong P, et al. Convolutional recurrent neural network for dynamic functional MRI analysis and brain ...

    work page 2022

  3. [13]

    BolT: Fused window transformers for fMRI time series analysis[J]

    Bedel H A, Sivgin I, Dalmaz O, et al. BolT: Fused window transformers for fMRI time series analysis[J]. Medical image analysis, 2023, 88: 102841. [14] Yu X, Zhang L, Cao C, et al. Gyri vs. Sulci: Core-Periphery Organization in Functional Brain Networks[C]//International Conference on Medical Image Computing and Computer-Assisted Intervention. Cham: Spring...

    work page 2023

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.