Pith. sign in

REVIEW 2 major objections 1 minor 47 references

GCAN generates counterfactual connectomes from multimodal brain data to classify cognitive decline stages while producing attention maps of connectivity changes.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · grok-4.3

2026-06-28 17:05 UTC pith:I5GBHKC2

load-bearing objection The paper introduces a generative counterfactual attention setup with an atlas-aware transformer for multimodal connectomes, but the abstract supplies no performance numbers or fidelity metrics to support the claims. the 2 major comments →

arxiv 2606.01237 v1 pith:I5GBHKC2 submitted 2026-05-31 cs.AI

Brain-Atlas-Guided Generative Counterfactual Attention for Explainable Cognitive Decline Diagnosis Using Multimodal Connectomes

Xiongri Shen , Jiaqi Wang , Zhenxi Song , Yi Zhong , Leilei Zhao , Xin He , Baiying Lei , Zhiguo Zhang This is my paper
classification cs.AI
keywords cognitive declinemultimodal connectomescounterfactual generationattention mapsbrain atlasfunctional connectivitystructural connectivityAlzheimer's disease
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper introduces an atlas-knowledge-guided generative counterfactual attention network called GCAN that treats diagnosis of subjective cognitive decline and mild cognitive impairment as a problem of generating target-state connectomes from source inputs. Differences between generated and input connectomes yield attention maps that highlight functional reorganization and structural topology changes. The method uses an atlas-aware bidirectional transformer to maintain network topology and extends from functional to joint functional-structural modeling. Experiments on hospital and ADNI datasets report competitive classification across healthy control versus SCD, healthy control versus MCI, and SCD versus MCI tasks. A sympathetic reader would care because the approach aims to combine diagnostic performance with visual explanations of disease-related connectivity alterations.

Core claim

GCAN formulates diagnosis as a source-to-target counterfactual generation problem, where target-label connectomes are generated from source-label inputs and their differences are used to construct counterfactual attention maps. An Atlas-aware Bidirectional Transformer performs network-level token encoding and decoding under brain-atlas constraints to preserve connectome topology. The framework extends to joint functional and structural connectivity modeling, with modality-specific pre-trained classifiers supplying target-state priors while remaining separate from the downstream diagnostic classifier.

What carries the argument

Generative Counterfactual Attention-guided Network (GCAN) that generates target-label connectomes and derives attention maps from their differences, guided by an Atlas-aware Bidirectional Transformer (AABT) for topology-preserving encoding and decoding.

Load-bearing premise

The generative model can produce realistic target-label connectomes that reflect actual disease-related connectivity changes without artifacts.

What would settle it

A direct comparison showing that the generated counterfactual connectomes have connectivity patterns statistically indistinguishable from real target-group data would support the claim; systematic mismatch in edge weights or failure of attention maps to align with known Alzheimer's connectivity signatures would refute it.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • Competitive accuracy is achieved on HC versus SCD, HC versus MCI, and SCD versus MCI tasks on both hospital-collected and ADNI datasets.
  • Counterfactual attention maps provide visualizations of functional and structural connectivity differences between diagnostic groups.
  • Joint modeling of functional and structural connectomes enables analysis of complementary reorganization patterns.
  • Separation of pre-trained modality classifiers from the diagnostic classifier avoids data leakage while supplying generation priors.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • If the generated maps reliably isolate disease effects, they could be tested against longitudinal patient trajectories to see whether highlighted regions predict conversion rates.
  • The atlas-constrained generation approach might be adapted to other neurodegenerative conditions where connectome topology changes are central.
  • Clinicians could compare the attention maps against established pathology atlases to check for overlap with amyloid or tau deposition patterns.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 1 minor

Summary. The manuscript proposes GCAN, an atlas-knowledge-guided Generative Counterfactual Attention-guided Network for explainable diagnosis of cognitive decline (HC vs. SCD, HC vs. MCI, SCD vs. MCI) from multimodal brain connectomes. Diagnosis is framed as source-to-target counterfactual generation: an Atlas-aware Bidirectional Transformer (AABT) generates target-label connectomes while preserving topology under brain-atlas constraints; differences yield counterfactual attention maps. The framework extends to joint FC/SC modeling, uses separated pre-trained modality-specific classifiers for target priors to avoid leakage, and reports competitive performance on hospital-collected and ADNI datasets, supported by visualizations, circular connectome analysis, CAM comparisons, ablations, and confidence intervals.

Significance. If the generated counterfactuals accurately capture disease-related connectivity alterations, the approach could advance interpretable multimodal connectomics by linking model decisions to specific functional reorganization and structural topology changes in the early AD continuum. Strengths include the atlas-aware tokenization for topology preservation, explicit separation of pre-trained classifiers to reduce leakage risk, and inclusion of ablations plus confidence interval analysis. These elements support reliability claims if the core generative fidelity holds.

major comments (2)
  1. [Abstract] Abstract: the claim that 'GCAN achieves competitive performance across HC vs. SCD, HC vs. MCI, and SCD vs. MCI classification tasks' is unsupported by any numerical results, baseline comparisons, AUC/accuracy values, or error bars, which is load-bearing for the central effectiveness assertion.
  2. [Methods] Methods (counterfactual generation and AABT description): no quantitative fidelity metrics (e.g., MMD, Wasserstein distance, or correlation with known AD biomarkers such as DMN reorganization) are reported to confirm that source-to-target generated connectomes reflect actual disease changes rather than model artifacts; this directly undermines the interpretability claim that attention maps correspond to clinically meaningful alterations.
minor comments (1)
  1. [Abstract] Abstract: the phrasing 'further support the interpretability and reliability' is vague without referencing specific figures or tables; adding one or two key quantitative highlights from the results would improve clarity.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback and positive assessment of the framework's strengths, including atlas-aware tokenization and leakage prevention. We address each major comment below and will revise the manuscript to improve clarity and support for the claims.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the claim that 'GCAN achieves competitive performance across HC vs. SCD, HC vs. MCI, and SCD vs. MCI classification tasks' is unsupported by any numerical results, baseline comparisons, AUC/accuracy values, or error bars, which is load-bearing for the central effectiveness assertion.

    Authors: We agree that the abstract would benefit from explicit numerical support. The full manuscript reports AUC, accuracy, and confidence intervals with baseline comparisons in the results section, but these were omitted from the abstract for brevity. In the revised version, we will incorporate key metrics (e.g., AUC values with CIs and comparisons) directly into the abstract to substantiate the performance claim. revision: yes

  2. Referee: [Methods] Methods (counterfactual generation and AABT description): no quantitative fidelity metrics (e.g., MMD, Wasserstein distance, or correlation with known AD biomarkers such as DMN reorganization) are reported to confirm that source-to-target generated connectomes reflect actual disease changes rather than model artifacts; this directly undermines the interpretability claim that attention maps correspond to clinically meaningful alterations.

    Authors: This is a fair point regarding validation of the generative component. The manuscript relies on visualizations, circular connectome analysis, CAM comparisons, and ablations to support that attention maps reflect meaningful changes, but does not include quantitative fidelity metrics. We will add MMD, Wasserstein distance, and correlations with AD biomarkers (e.g., DMN) to the revised methods/results to provide stronger quantitative evidence for the counterfactuals. revision: yes

Circularity Check

0 steps flagged

No significant circularity in derivation chain

full rationale

The paper presents GCAN as an empirical generative architecture for counterfactual attention maps from multimodal connectomes, using pre-trained modality-specific classifiers explicitly separated from the diagnostic classifier to avoid leakage. No equations or steps in the abstract reduce by construction to inputs (e.g., no self-definitional generation where outputs are fitted parameters renamed as predictions, no load-bearing self-citations, and no uniqueness theorems imported from prior author work). The framework is a proposed DL model evaluated on external datasets (hospital and ADNI), making the derivation self-contained rather than tautological. The interpretability claim rests on experimental visualizations rather than definitional equivalence.

Axiom & Free-Parameter Ledger

1 free parameters · 2 axioms · 0 invented entities

The framework rests on several domain assumptions about connectome topology preservation and counterfactual validity plus numerous learned parameters typical of transformer-based generative models; no independent evidence for invented entities is provided.

free parameters (1)
  • model hyperparameters and training parameters
    Deep learning components including transformer layers, attention mechanisms, and generative losses require fitting to data.
axioms (2)
  • domain assumption Brain atlas provides valid constraints that preserve connectome topology during token encoding and decoding.
    Invoked in the Atlas-aware Bidirectional Transformer (AABT) description.
  • domain assumption Counterfactual generation from source to target label produces differences that reflect real disease-related connectivity changes.
    Central to constructing counterfactual attention maps.

pith-pipeline@v0.9.1-grok · 5823 in / 1245 out tokens · 21951 ms · 2026-06-28T17:05:41.425521+00:00 · methodology

0 comments
read the original abstract

Mild cognitive impairment (MCI) and subjective cognitive decline (SCD) are closely associated with the early Alzheimer's disease continuum, where accurate and explainable diagnosis is important for early risk assessment and intervention. Existing connectome-based deep learning models can improve classification performance but often provide limited insight into disease-related functional and structural connectivity changes. This paper proposes an atlas-knowledge-guided Generative Counterfactual Attention-guided Network (GCAN) for explainable cognitive decline diagnosis using multimodal brain connectomes. GCAN formulates diagnosis as a source-to-target counterfactual generation problem, where target-label connectomes are generated from source-label inputs and their differences are used to construct counterfactual attention maps. To preserve connectome topology, an Atlas-aware Bidirectional Transformer (AABT) performs network-level token encoding and decoding under brain-atlas constraints. The framework is further extended from functional connectivity (FC) to joint functional and structural connectivity (SC) modeling, enabling counterfactual analysis of complementary functional reorganization and structural topology changes. Experiments on hospital-collected and ADNI datasets show that GCAN achieves competitive performance across HC vs. SCD, HC vs. MCI, and SCD vs. MCI classification tasks. Visualization, circular connectome analysis, CAM-based comparison, ablation studies, and confidence interval analysis further support the interpretability and reliability of the proposed framework. Modality-specific FC and SC pre-trained classifiers are used to provide target-state priors for counterfactual generation while being separated from the downstream diagnostic classifier to prevent data leakage.

Figures

Figures reproduced from arXiv: 2606.01237 by Baiying Lei, Jiaqi Wang, Leilei Zhao, Xin He, Xiongri Shen, Yi Zhong, Zhenxi Song, Zhiguo Zhang.

Figure 1
Figure 1. Figure 1: Overview of the proposed GCAN framework for counterfactual attention-guided cognitive decline diagnosis. During training, GCAN generates target-label functional connectivity from source-label functional connectivity and derives positive and negative counterfactual attention maps by measuring their differences. During prediction, the learned counterfactual attention is applied to the input functional connec… view at source ↗
Figure 2
Figure 2. Figure 2: Generator–discriminator architecture of the single-modal GCAN. The generator reconstructs the source-label functional connectivity and transforms it into target-label functional connectivity by incorporating target-state information. The discriminator contains image-level and neurodegeneration-level components to constrain the realism and class￾discriminative properties of the generated connectomes. 3.3. G… view at source ↗
Figure 3
Figure 3. Figure 3: Architecture of the AABT for functional connectivity modeling. AABT decomposes the input functional connectivity or feature map into atlas-defined brain networks, performs network-specific patch and position embedding, and then reconstructs the output through inverse patch embedding. This bidirectional encoding–decoding process preserves the structured topology of brain connectomes. where 𝑃𝑘 (⋅) denotes th… view at source ↗
Figure 4
Figure 4. Figure 4: Overview of the multimodal GCAN framework for joint functional and structural connectome analysis. In the training stage, source-label FC/SC connectomes are transformed into target-label FC/SC connectomes to construct multimodal counterfactual attention. In the prediction stage, positive and negative attention maps are aggregated and applied to multimodal connectomes to support explainable cognitive declin… view at source ↗
Figure 5
Figure 5. Figure 5: Architecture of the cross-modal AABT for joint FC/SC modeling. The input FC, SC, or feature maps are segmented according to atlas-defined brain networks. Network-specific tokens are generated through patch and position embedding, processed by self-attention and feed-forward layers, and then reconstructed through inverse patch embedding to preserve cross-modal connectome topology. 3.6. Multimodal optimizati… view at source ↗
Figure 6
Figure 6. Figure 6: Generator–discriminator architecture of the multimodal GCAN. The generator learns source-to-target transforma￾tions between multimodal FC/SC connectomes, while the discriminator imposes image-level and neurodegeneration-level constraints on the generated connectomes. Pretrained classifiers provide label-aware supervision for source and target cognitive states. where 𝐴𝑚 is the bidirectional counterfactual a… view at source ↗
Figure 7
Figure 7. Figure 7: Visualization of single-modal counterfactual attention maps across different cognitive-state transition tasks. For each task, the generated target-label FC is compared with the source-label FC, and their difference is projected onto brain regions to obtain positive and negative counterfactual attention maps. The highlighted regions indicate discriminative brain areas associated with functional connectivity… view at source ↗
Figure 8
Figure 8. Figure 8: Circular connectome visualization of single-modal FC counterfactual analysis across different cognitive-state transition tasks. Source-label FC, generated target-label FC, and counterfactual attention are shown in three columns. Red and blue connections indicate increased and decreased connectivity, respectively. differences between source and target states are not uniformly distributed across the brain. I… view at source ↗
Figure 9
Figure 9. Figure 9: Visualization of multimodal counterfactual attention maps across different cognitive-state transition tasks. The upper and lower panels show bidirectional transitions among HC, SCD, and MCI. For each transition, the source-label FC/SC matrices and generated target-label FC/SC matrices are compared to obtain positive and negative counterfactual attention maps. subjective decline to mild impairment is associ… view at source ↗
Figure 10
Figure 10. Figure 10: Multimodal circular connectome visualization of FC and SC counterfactual analysis across different cognitive￾state transition tasks. The upper panel shows functional connectivity (FC), and the lower panel shows structural connectivity (SC). For each modality, source-label connectomes, generated target-label connectomes, and counterfactual attention maps are shown from left to right. Red and purple/blue co… view at source ↗
Figure 11
Figure 11. Figure 11: Statistical comparison of multimodal diagnostic performance across three classification tasks. Boxplots show the fold-wise distributions of ACC, recall, precision, and F1-score. Yellow boxes indicate the selected best attention model, red boxes indicate literature-mapped baseline methods, and cyan boxes indicate counterfactual-attention-based methods. ∗ denotes using counterfactual attention. Statistical … view at source ↗
Figure 12
Figure 12. Figure 12: Ablation visualization of AABT on multimodal circular connectome attention maps. The six columns correspond to six source-to-target transition directions. The four rows show FC attention with AABT, FC attention without AABT, SC attention with AABT, and SC attention without AABT, respectively. The visualization compares whether atlas-aware token encoding and decoding produce more structured network-level c… view at source ↗
Figure 13
Figure 13. Figure 13: Comparison of circular connectome visualizations between the proposed counterfactual attention and CAM￾based interpretation methods. Rows denote six cognitive-state transitions, and columns denote FC and SC modalities with the proposed method, Grad-CAM, and Score-CAM. The proposed method highlights more transition-specific and signed connectivity changes, whereas CAM-based methods mainly provide classifie… view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

47 extracted references

  1. [1]

    Computers in Biology and Medicine , volume=

    A spatiotemporal graph transformer approach for Alzheimer's disease diagnosis with rs-fMRI , author=. Computers in Biology and Medicine , volume=. 2024 , doi=

    2024

  2. [2]

    2025 , eprint=

    Classification of Mild Cognitive Impairment Based on Dynamic Functional Connectivity Using Spatio-Temporal Transformer , author=. 2025 , eprint=

    2025

  3. [3]

    Medical Image Computing and Computer Assisted Intervention -- MICCAI 2025 , volume=

    Cross-Modal Brain Graph Transformer via Function-Structure Connectivity Network for Brain Disease Diagnosis , author=. Medical Image Computing and Computer Assisted Intervention -- MICCAI 2025 , volume=. 2025 , publisher=

    2025

  4. [4]

    IEEE Transactions on Neural Systems and Rehabilitation Engineering , volume=

    Alzheimer's Disease Prediction via Brain Structural-Functional Deep Fusing Network , author=. IEEE Transactions on Neural Systems and Rehabilitation Engineering , volume=. 2023 , doi=

    2023

  5. [5]

    Computer Methods and Programs in Biomedicine , volume=

    An extensible hierarchical graph convolutional network for early Alzheimer's disease identification , author=. Computer Methods and Programs in Biomedicine , volume=. 2023 , doi=

    2023

  6. [6]

    IEEE Transactions on Neural Systems and Rehabilitation Engineering , volume=

    Brain Structure-Function Fusing Representation Learning Using Adversarial Decomposed-VAE for Analyzing MCI , author=. IEEE Transactions on Neural Systems and Rehabilitation Engineering , volume=. 2023 , doi=

    2023

  7. [7]

    IEEE Transactions on Cybernetics , volume=

    Prior-Guided Adversarial Learning With Hypergraph for Predicting Abnormal Connections in Alzheimer's Disease , author=. IEEE Transactions on Cybernetics , volume=. 2024 , doi=

    2024

  8. [8]

    Transformer Graph Convolutional Networks For Accurate Classification Of

    Huang, Maochang and Liu, Tao and Chen, Huijuan and Huang, Weiyuan and Lv, Chaoqi and Zang, Lan and Shen, Chong and Guo, Yihao and Chen, Feng , journal=. Transformer Graph Convolutional Networks For Accurate Classification Of. 2026 , publisher=

    2026

  9. [9]

    Design Of A

    Velu, K and Jaisankar, N , journal=. Design Of A. 2025 , publisher=

    2025

  10. [10]

    Vision Transformers For The Prediction Of Mild Cognitive Impairment To

    Hoang, Gia Minh and Kim, Ue-Hwan and Kim, Jae Gwan , journal=. Vision Transformers For The Prediction Of Mild Cognitive Impairment To. 2023 , publisher=

    2023

  11. [11]

    Understanding The Role Of Self-Attention In A Transformer Model For The Discrimination Of

    Sibilano, Elena and Buongiorno, Domenico and Lassi, Michael and Grippo, Antonello and Bessi, Valentina and Sorbi, Sandro and Mazzoni, Alberto and Bevilacqua, Vitoantonio and Brunetti, Antonio , journal=. Understanding The Role Of Self-Attention In A Transformer Model For The Discrimination Of. 2024 , publisher=

    2024

  12. [12]

    Identification Of Mild Cognitive Impairment Based On Quadruple

    Li, Yuan and Zou, Ying and Guo, Hanning and Yang, Yongqing and Li, Na and Li, Linhao and Zhao, Feng , journal=. Identification Of Mild Cognitive Impairment Based On Quadruple. 2023 , publisher=

    2023

  13. [13]

    Multimodal Classification Of

    Adarsh, V and Gangadharan, GR and Fiore, Ugo and Zanetti, Paolo , journal=. Multimodal Classification Of. 2024 , publisher=

    2024

  14. [14]

    Multimodal Diagnosis Model Of

    Tang, Yan and Xiong, Xing and Tong, Gan and Yang, Yuan and Zhang, Hao , journal=. Multimodal Diagnosis Model Of. 2024 , publisher=

    2024

  15. [15]

    Multi-Modal Medical Image Fusion Based On Densely-Connected High-Resolution

    Zhou, Quan and Ye, Shaozhuang and Wen, Mingwei and Huang, Zhiwen and Ding, Mingyue and Zhang, Xuming , journal=. Multi-Modal Medical Image Fusion Based On Densely-Connected High-Resolution. 2022 , publisher=

    2022

  16. [16]

    Structural

    Kang, Xiaopeng and Lin, Jiaji and Zhao, Kun and Yan, Shaozhen and Chen, Pindong and Wang, Dawei and Yao, Hongxiang and Zhou, Bo and Yu, Chunshui and Wang, Pan and others , journal=. Structural. 2025 , publisher=

    2025

  17. [17]

    Multi-Modal Feature Fusion Using Transformer For Early

    Li, Jiaqiang and Gao, Yian and Yang, Peng and Guan, Zhenghua and Wang, Tianfu and Ma, Guolin and Lei, Baiying , booktitle=. Multi-Modal Feature Fusion Using Transformer For Early. 2025 , address=

    2025

  18. [18]

    Early Prediction Of Sudden Cardiac Death Using Multimodal Fusion Of

    Telangore, Hardik and Azad, Victor and Sharma, Manish and Bhurane, Ankit and San Tan, Ru and Acharya, U Rajendra , journal=. Early Prediction Of Sudden Cardiac Death Using Multimodal Fusion Of. 2024 , publisher=

    2024

  19. [19]

    Tasji, Linah and Alghamdi, Hanan S and Almalaise Al-Ghamdi, Abdullah S , journal=. Toward. 2026 , publisher=

    2026

  20. [20]

    Optimizing Pretrained

    Prasetyo, Simeon Yuda , booktitle=. Optimizing Pretrained. 2024 , address=

    2024

  21. [21]

    A Deep Learning Approach For Automated Diagnosis And Multi-Class Classification Of

    Ramzan, Farheen and Khan, Muhammad Usman Ghani and Rehmat, Asim and Iqbal, Sajid and Saba, Tanzila and Rehman, Amjad and Mehmood, Zahid , journal=. A Deep Learning Approach For Automated Diagnosis And Multi-Class Classification Of. 2020 , publisher=

    2020

  22. [22]

    Analysis Of Features Of

    Odusami, Modupe and Maskeli. Analysis Of Features Of. Diagnostics , volume=. 2021 , publisher=

    2021

  23. [23]

    A Transfer Convolutional Neural Network For Fault Diagnosis Based On

    Wen, Long and Li, Xinyu and Gao, Liang , journal=. A Transfer Convolutional Neural Network For Fault Diagnosis Based On. 2020 , publisher=

    2020

  24. [24]

    Cerebral Cortex , volume=

    Functional Connectivity Hypersynchronization in Relatives of Alzheimer’s Disease Patients: An Early E/I Balance Dysfunction? , author=. Cerebral Cortex , volume=. 2021 , publisher=

    2021

  25. [25]

    Human Brain Mapping , volume=

    Investigation of the Functional Pathogenesis of Mild Cognitive Impairment by Localisation-Based Locus Coeruleus Resting-State fMRI , author=. Human Brain Mapping , volume=. 2022 , publisher=

    2022

  26. [26]

    IEEE Transactions on Medical Imaging , volume=

    Virtual Adversarial Training-Based Deep Feature Aggregation Network from Dynamic Effective Connectivity for MCI Identification , author=. IEEE Transactions on Medical Imaging , volume=. 2021 , publisher=

    2021

  27. [27]

    Frontiers in Computational Neuroscience , volume=

    U-Shaped Convolutional Transformer GAN with Multi-Resolution Consistency Loss for Restoring Brain Functional Time-Series and Dementia Diagnosis , author=. Frontiers in Computational Neuroscience , volume=. 2024 , publisher=

    2024

  28. [28]

    Proceedings of the IEEE International Conference on Computer Vision , address=

    Grad-Cam: Visual Explanations from Deep Networks via Gradient-Based Localization , author=. Proceedings of the IEEE International Conference on Computer Vision , address=

  29. [29]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops , address=

    Score-CAM: Score-Weighted Visual Explanations for Convolutional Neural Networks , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops , address=

  30. [30]

    IEEE Transactions on Pattern Analysis and Machine Intelligence , volume=

    Learn-Explain-Reinforce: Counterfactual Reasoning and Its Guidance to Reinforce an Alzheimer's Disease Diagnosis Model , author=. IEEE Transactions on Pattern Analysis and Machine Intelligence , volume=. 2022 , publisher=

    2022

  31. [31]

    International Conference on Medical Image Computing and Computer-Assisted Intervention , address=

    Punctate White Matter Lesion Segmentation in Preterm Infants Powered by Counterfactually Generative Learning , author=. International Conference on Medical Image Computing and Computer-Assisted Intervention , address=. 2023 , organization=

    2023

  32. [32]

    Archives of Neurology , volume=

    Mild cognitive impairment: clinical characterization and outcome , author=. Archives of Neurology , volume=. 1999 , doi=

    1999

  33. [33]

    Alzheimer's & Dementia , volume=

    A conceptual framework for research on subjective cognitive decline in preclinical Alzheimer's disease , author=. Alzheimer's & Dementia , volume=. 2014 , doi=

    2014

  34. [34]

    Frontiers in Systems Neuroscience , volume=

    Clinical applications of resting state functional connectivity , author=. Frontiers in Systems Neuroscience , volume=. 2010 , doi=

    2010

  35. [35]

    Nature Reviews Neuroscience , volume=

    Complex brain networks: graph theoretical analysis of structural and functional systems , author=. Nature Reviews Neuroscience , volume=. 2009 , doi=

    2009

  36. [36]

    Nature Machine Intelligence , volume=

    Stop explaining black box machine learning models for high stakes decisions and use interpretable models instead , author=. Nature Machine Intelligence , volume=. 2019 , doi=

    2019

  37. [37]

    Harvard Journal of Law & Technology , volume=

    Counterfactual explanations without opening the black box: Automated decisions and the GDPR , author=. Harvard Journal of Law & Technology , volume=. 2018 , doi=

    2018

  38. [38]

    Medical Image Analysis , volume=

    Interpretable modality-specific and interactive graph convolutional network on brain functional and structural connectomes , author=. Medical Image Analysis , volume=. 2025 , doi=

    2025

  39. [39]

    Neural Networks , volume=

    Explainable multiplex graph propagational network with multimodal neuroimage integration for dementia subtype diagnosis , author=. Neural Networks , volume=. 2025 , doi=

    2025

  40. [40]

    npj Artificial Intelligence , volume=

    Rethinking functional brain connectome analysis: do graph deep learning models help , author=. npj Artificial Intelligence , volume=. 2026 , doi=

    2026

  41. [41]

    Diagnostics , volume=

    Explainable artificial intelligence in neuroimaging of Alzheimer's disease , author=. Diagnostics , volume=. 2025 , doi=

    2025

  42. [42]

    NeuroImage , volume=

    Local structural--functional coupling with counterfactual explanations for epilepsy prediction , author=. NeuroImage , volume=. 2025 , doi=

    2025

  43. [43]

    MIT Press , year=

    Networks of the brain , author=. MIT Press , year=

  44. [44]

    European Conference on Computer Vision , address=

    Cbam: Convolutional Block Attention Module , author=. European Conference on Computer Vision , address=

  45. [45]

    Alzheimer's & Dementia , volume=

    Distinct Patterns of Posterior Default Mode Network-Medial Temporal Lobe Connectivity in Mild Cognitive Impairment and Subjective Cognitive Decline , author=. Alzheimer's & Dementia , volume=. 2021 , publisher=

    2021

  46. [46]

    Human Brain Mapping , volume=

    Accumulation of Network Redundancy Marks the Early Stage of Alzheimer's Disease , author=. Human Brain Mapping , volume=. 2023 , publisher=

    2023

  47. [47]

    IEEE Journal of Biomedical and Health Informatics , volume=

    A Hybrid Multi-Scale Attention Convolution and Aging Transformer Network for Alzheimer's Disease Diagnosis , author=. IEEE Journal of Biomedical and Health Informatics , volume=. 2023 , publisher=

    2023