Pith. sign in

REVIEW 4 major objections 5 minor 33 references

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

T0 review · 4 major / 5 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read Treating brain subnetworks as tokens aligned with clinical report words yields 92.94% accuracy in distinguishing mild cognitive impairment from controls and recovers tau-related connectome-text associations.

desk verdict The architecture is a plausible novelty, but the evaluation leaks the label through the text, so the headline numbers are not credible as written. read the letter →

arxiv 2508.06565 v1 pith:2KF2E4WE submitted 2025-08-07 cs.CV cs.LG

Jing Zhang , Xiaowei Yu , Minheng Chen , Lu Zhang , Tong Chen , Yan Zhuang , Chao Cao , Yanjun Lyu
show 3 more authors
Li Su Tianming Liu Dajiang Zhu
This is my paper
classification cs.CVcs.LG
keywords medicalvision-languagemodelmultimodallearningmildcognitiveimpairmentbrainconnectomeDTIcross-modalalignmentAlzheimer'sdiseaseinterpretability
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 claims that combining a DTI brain connectome with a generated clinical narrative can diagnose mild cognitive impairment more accurately than imaging-only models, while also producing interpretable links between brain networks and clinical vocabulary. Its central move is to treat each brain region's connectivity profile as a token, like a word or image patch, and align those tokens with BERT word tokens from the clinical text in a shared latent space at two levels: whole-subject and fine-grained connectome-level. The authors report 92.94% accuracy on MCI versus normal controls, with ablations showing that removing either modality or either alignment level lowers performance, and they present recovered pairs such as phosphorylated_tau with cingulate subnetworks as evidence that the alignments are clinically meaningful. If correct, this gives clinicians a way to turn routine reports and a single diffusion scan into both a prediction and a map of which text observations go with which brain systems.

What carries the argument

The load-bearing object is the brain sub-network token: the row of the structural-connectivity matrix that records one brain region's fiber connectivity to every other region, making network-level rather than voxel-level information the unit of alignment. A vanilla ViT encoder treats these tokens like image patches; a BERT encoder treats clinical-report words as tokens; and bidirectional cross-attention between the two token sets produces similarity scores that are trained with weighted softmax contrastive losses at connectome level and InfoNCE-style losses at subject level. The category-level claim is carried by this two-level alignment plus the classification head.

What would settle it

Retrain the model on the same reports with all cognitive-scale values (MMSE, CDR) and outcome-defining biomarker fields removed, keeping only the doctors' narrative wording. If accuracy falls to roughly the imaging-only level (73.77%) or the text-only advantage disappears, the cross-modal claim is not supported; if it stays near 92.94%, the alignment is genuinely using independent clinical semantics.

Watch

Extended reading notes

Core claim

The discovery claimed is that brain subnetworks can serve as the imaging-side tokens in a vision-language alignment, and that aligning them with clinical-report word tokens yields better MCI classification and clinically interpretable pairs. From the structural connectivity matrix $SC \in \mathbb{R}^{N\times N}$, each row is treated as a sub-network token; a learnable class token aggregates all rows, and the sequence is fed to a vanilla ViT. A BERT encoder produces token- and report-level text representations. Bidirectional cross-attention between sub-network tokens and word tokens yields a connectome-level cosine-similarity matrix, while a subject-level InfoNCE-style loss matches global bra

Load-bearing premise

The load-bearing premise is that the clinical narratives do not already encode the diagnostic outcome, even though they are generated from MMSE and CDR scores that are partly used to define MCI, so the text and imaging are treated as independent evidence.

Editorial extensions

If this is right

  • If the accuracy claim holds, DTI plus a text report is enough to match or beat multi-imaging fusion, which lowers the acquisition burden for early AD screening.
  • The recovered connectome-text pairs (tau, aβ42, MMSE, CDR tokens mapped to cingulate and parietal subnetworks) give concrete region-versus-measure hypotheses that clinical studies could test prospectively.
  • Treating connectome rows as tokens offers a general recipe: the same alignment machinery can be applied to functional connectivity, morphometry, or PET-derived networks with the same report encoder.
  • Ablation results suggest both alignment levels matter, meaning fine-grained subnetwork-to-word correspondences carry information beyond whole-subject matching.

Reading between the lines

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

  • The paper does not isolate whether the clinical narrative's cognitive scores are necessary for the gain; an extension would retrain with reports stripped of MMSE, CDR, and biomarker fields to measure how much of the 92.94% comes from genuinely independent text semantics.
  • If the connectome-token recipe ports to other modalities, the framework could be used to mine large report repositories for disease-specific network terminology without manual annotation.
  • The reported tau-cingulate and aβ42-parietal pairs could be treated as priors for targeted ROI analyses, effectively turning the model's attention weights into hypotheses about where early pathology appears before atrophy is visible.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The paper proposes a multimodal framework for MCI diagnosis that aligns DTI-derived structural connectomes with clinical report text. Brain subnetworks (rows of the SC matrix) are treated as tokens and encoded by a ViT; clinical reports are encoded by BERT; cross-modal attention is applied at both the subject level and the connectome level. On ADNI data (301 NC, 117 MCI), the method reports 92.94% accuracy, with text-only and imaging-only ablations at 86.96% and 73.77%, respectively. The paper also presents visualizations of top brain-subnetwork/text-token pairs and claims state-of-the-art performance and clinically meaningful multimodal associations.

Significance. If the reported results were valid, the paper would make a useful contribution: the idea of treating brain subnetworks as tokens for connectome-text alignment is a sensible inductive bias, the method is clearly described, and the code is stated to be public. However, the central evidence is undermined by a likely label-leakage problem. Section 3.1 states that clinical narratives are generated from demographic information, biomarker measurements (APOE-4), cognitive assessments (MMSE, CDR), and doctors' notes. Since MCI status in ADNI is operationally defined in part by MMSE and CDR scores, the text modality can encode the outcome before imaging is used. The 86.96% text-only accuracy in Table 2 is close to the 92.94% multimodal accuracy, suggesting the multimodal gain may be a small increment on a label-leaking text feature rather than meaningful cross-modal fusion. The interpretability findings in Figs. 3-4 are also largely circular because tokens such as phosphorylated_tau, aβ42, hippocampus, and entorhinal_cortex are copied from the same structured fields that define the label. These concerns affect the paper's two central claims, so the work currently does not support its stated

major comments (4)
  1. [Section 3.1, Table 2] Label leakage: the text modality is assembled from MMSE, CDR, APOE-4, and CSF biomarker values, which are the same variables used to define the MCI/NC label in ADNI. The text-only C-only row reaches 86.96% accuracy, only 6 percentage points below the multimodal result. Under these conditions, the 92.94% accuracy and the claimed benefit of cross-modal alignment may simply reflect the model reading the diagnosis from structured fields inserted into the report. Please rerun the experiments with reports redacted to remove all label-proximate variables (or at least show that text-only performance on redacted doctors' notes is near chance), and report whether the multimodal advantage persists.
  2. [Section 3.2-3.3, Figs. 3-4] The 'clinically meaningful connectome-text pairs' are largely circular. Tokens such as phosphorylated_tau, aβ42, hippocampus, entorhinal_cortex, MMSE, and CDR are field names or structured values that the authors explicitly wrote into the clinical narratives. Recovering associations between these tokens and known AD-related subnetworks is expected by construction and does not provide independent evidence that the model learned medical knowledge. Please add control analyses: for example, permute token-to-subnetwork assignments, remove all biomarker/cognitive phrases, or compare against a bag-of-words model trained on the same narratives. Without such controls, the interpretability claims in the abstract and conclusion are unsupported.
  3. [Section 3.4, Table 1] The state-of-the-art claim is not supported by the comparison in Table 1. The cited studies use different datasets, cohort sizes, modalities, and evaluation splits; no statistical significance tests or confidence intervals are reported. The 92.94% accuracy on 84 test subjects (20% of 418) has a wide confidence interval, so the comparison is not meaningful. Please include same-data baselines under identical train/test splits, repeated cross-validation, and confidence intervals or significance tests.
  4. [Section 3.1, Table 2] The experimental protocol is not sufficiently robust for the strength of the claims. A single 80/20 split with 418 subjects yields a test set of roughly 84 subjects; the difference between 90.59% and 92.94% in the alignment ablation is within sampling noise. There is no mention of repeated splits, seeds, class-balancing details, or external validation. Please report mean and standard deviation over multiple runs or cross-validation folds, and perform a paired significance test for the ablation comparisons.
minor comments (5)
  1. [Table 1] Reference numbering is inconsistent: reference [16] is listed as both 'Zhang et al. (2025) rsfMRI' and 'Song et al. (2024) DTI+record'. The row in category (C) labeled 'Song et al. (2024) [16]' appears to be a citation error.
  2. [Table 2 caption] The symbols ℒ!" and ℒ#" in the caption are garbled; the text uses different notation for connectome-level and subject-level losses. Please standardize the notation.
  3. [Section 2.2] Several loss equations have malformed parentheses, e.g., L_ct2i and L_ci2t contain unmatched parentheses and an unusual sum over softmax terms. The formulas should be rewritten in standard notation.
  4. [Figure 2] The figure is overcrowded and several labels are difficult to read, especially in the clinical report encoder and the cross-attention blocks. Please enlarge fonts and separate the alignment schematic from the architecture diagram.
  5. [Section 3.1] The statement 'Code is publicly available' has no URL. If the code is available, please include a link; if not, remove the sentence.

Circularity Check

2 steps flagged · score 6.0 of 10

Text modality is constructed from label-defining variables, so the reported accuracy and connectome-text 'findings' reduce to label leakage and input re-description rather than independent cross-modal prediction.

  1. self definitional [Section 3.1, Data Preprocessing and Experimental Design; Table 2(A)]
    "The non-imaging dataset encompassed comprehensive patient information including demographic information (e.g., age, sex, education level), biomarker measurements (e.g., APOE-4), cognitive assessments (e.g., MMSE, CDR), and unstructured doctors' clinical notes. These components were integrated to generate comprehensive clinical narratives for each patient."

    The clinical narratives used as the text input are generated from MMSE, CDR, and biomarker measurements. MCI status in ADNI is operationally defined in part by these same cognitive assessments and biomarkers, so the text modality encodes the target label before imaging is used. Table 2 shows C-only reaches 86.96% accuracy, close to the 92.94% multimodal result, confirming that the text alone is nearly sufficient for the label. The paper's claim of 'complementary' multimodal information is therefore not established; the text input already contains the outcome by construction.

  2. renaming known result [Section 3.3, Key clinical text tokens; Fig. 4]
    "Fig. 4 presents the top-5 MCI-related text tokens extracted from the clinical reports and their corresponding significant Text-to-Brain interactions. ... A notable example is the strong association between the token phosphorylated_tau and cingulate brain regions ... Our model successfully captures this relationship, identifying brain sub-networks G_and_S_cingul-Mid-Post and G_cingul-Post-dorsal through text analysis, which closely aligns with established neuroimaging findings."

    The 'discovered' tokens—phosphorylated_tau, aβ42, hippocampus, entorhinal_cortex, cerebrospinal_fluid—are the same structured biomarker and region names that were assembled into the clinical narratives in Section 3.1. The attention between these tokens and brain subnetworks is therefore recovering the input's own diagnostic vocabulary, not producing an independent cross-modal prediction. Presenting this as a model 'finding' that 'validates' the framework renames known AD-biomarker/region associations as novel discoveries; the association is built into the text by construction.

full rationale

The core circularity is in the data construction, not in the network architecture or alignment losses. The clinical reports are explicitly built from MMSE, CDR, APOE-4, and other structured fields, and MCI is defined using those same clinical measures. Thus the text modality has direct access to the label, and the 86.96% text-only accuracy in Table 2 demonstrates the label is largely recoverable from the text alone. The multimodal gain over text-only is modest (92.94% vs 86.96%) and cannot be interpreted as evidence of independent cross-modal fusion. Similarly, the 'clinically meaningful connectome-text pairs' in Figs. 3-4 are dominated by tokens such as phosphorylated_tau, aβ42, MMSE, and CDR that were placed into the narrative by the authors; the attention mechanism is essentially retrieving the input variables that already carry diagnostic information, so calling these pairs a novel finding is a renaming of the known associations. The self-citations (e.g., refs. 3, 16, 23) are not load-bearing for the methodology or the quantitative results, so they do not add circularity. The alignment losses and transformer encoder are self-contained, but the paper's central claims of state-of-the-art prediction and new multimodal insights are partially circular because the text input is defined in terms of the target and the 'discovered' tokens are the input's own diagnostic vocabulary.

Assumptions & free parameters 6 free parameters · 4 assumptions · 0 invented entities

The central result relies on behavioral and biomarker information being placed in the text input, on the DTI connectome faithfully representing brain state, and on the subnetwork-as-token assumption. The first of these is likely false and drives the verdict; the others are domain assumptions without external validation. No new physical entities are introduced.

free parameters (6)
  • All trainable weights (patch embedding, ViT, BERT, cross-attention, classifier) = not reported
    Performance depends entirely on these learned weights; no pretraining or initialization details beyond BERT/ViT backbones.
  • Learnable [class] token SC_cls = not reported
    Introduced in Section 2.1 as a learned aggregation token for the connectome.
  • InfoNCE temperature tau = not stated
    Subject-level alignment loss in Eq. 1 contains an unspecified temperature/scale parameter that controls alignment tightness.
  • Hyperparameters (embedding dim, layers, epochs, batch, learning rate) = 256, 4, 32, 8, 1e-5
    Chosen by hand; no search or sensitivity analysis reported.
  • Loss balance weights / balancing strategy for CE loss = not stated
    The overall objective L = L_ca + L_sa + L_ce combines three losses; the 'balancing strategy' mentioned in Section 2.2 is not specified.
  • 80/20 train/test split assignment = not reported
    Single random split; no seed or stratification details, so results depend on split.
assumptions (4)
  • domain assumption ADNI MCI/NC labels are ground truth and are not encoded in the generated clinical narratives.
    Section 3.1 builds narratives from MMSE, CDR and biomarkers, which are part of MCI diagnosis; this assumption is likely false and is the main red flag.
  • domain assumption DTI structural connectivity matrix (fiber counts) is a valid representation of brain network state.
    Section 2.1 defines SC from DTI fiber counts via preprocessing from [4]; any loss of validity propagates to all imaging tokens.
  • ad hoc to paper Subnetwork rows of SC plus intra-modality self-attention capture disease-relevant network information.
    The claim that treating each region-to-all-others row as a token is better than patches is asserted, not derived or validated in isolation.
  • domain assumption BERT pretrained representations and cross-attention are appropriate for short structured clinical narratives.
    No domain adaptation or analysis of the text distribution is provided.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Bridging Brain Connectomes and Clinical Reports for Early Alzheimer's Disease Diagnosis." pith.science (2026). https://pith.science/paper/2KF2E4WE

@misc{pith2026250806565,
  author       = {Pith},
  title        = {Pith review of: Bridging Brain Connectomes and Clinical Reports for Early Alzheimer's Disease Diagnosis},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/2KF2E4WE}},
  note         = {Machine review of arXiv:2508.06565}
}
read the original abstract

Integrating brain imaging data with clinical reports offers a valuable opportunity to leverage complementary multimodal information for more effective and timely diagnosis in practical clinical settings. This approach has gained significant attention in brain disorder research, yet a key challenge remains: how to effectively link objective imaging data with subjective text-based reports, such as doctors' notes. In this work, we propose a novel framework that aligns brain connectomes with clinical reports in a shared cross-modal latent space at both the subject and connectome levels, thereby enhancing representation learning. The key innovation of our approach is that we treat brain subnetworks as tokens of imaging data, rather than raw image patches, to align with word tokens in clinical reports. This enables a more efficient identification of system-level associations between neuroimaging findings and clinical observations, which is critical since brain disorders often manifest as network-level abnormalities rather than isolated regional alterations. We applied our method to mild cognitive impairment (MCI) using the ADNI dataset. Our approach not only achieves state-of-the-art predictive performance but also identifies clinically meaningful connectome-text pairs, offering new insights into the early mechanisms of Alzheimer's disease and supporting the development of clinically useful multimodal biomarkers.

Figures

Figures reproduced from arXiv: 2508.06565 by the authors.

Figure 2
Figure 2. Fig.2. In [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

33 extracted references · 33 canonical work pages

  1. [1]

    ACC: Accuracy; SEN: Sensitivity; SPE: Specificity; F1: F1-score Method Modality NC/MCI ACC% SEN% SPE% F1% (A) Single-Imaging Modality Study Shao et al

    Comparison of identification performance with other recent state-of-the-art methods. ACC: Accuracy; SEN: Sensitivity; SPE: Specificity; F1: F1-score Method Modality NC/MCI ACC% SEN% SPE% F1% (A) Single-Imaging Modality Study Shao et al. (2020)

    work page 2020

  2. [3]

    MRI 225/120 69.14 79.34 55.92 - Song et al. (2024)

    work page 2024

  3. [4]

    Predicting brain structural network using functional con-nectivity[J]

    Zhang L, Wang L, Zhu D, et al. Predicting brain structural network using functional con-nectivity[J]. Medical image analysis, 2022, 79: 102463

    work page 2022

  4. [5]

    Parietal cortex matters in Alzheimer's disease: an overview of structural, functional and metabolic findings[J]

    Jacobs H I L, Van Boxtel M P J, Jolles J, et al. Parietal cortex matters in Alzheimer's disease: an overview of structural, functional and metabolic findings[J]. Neuroscience & Biobehav-ioral Reviews, 2012, 36(1): 297-309

    work page 2012

  5. [6]

    Cingulate cortex in pre-MCI cognition[J]

    Sambuchi N, Geda Y E, Michel B F. Cingulate cortex in pre-MCI cognition[J]. Handbook of clinical neurology, 2019, 166: 281-295

    work page 2019

  6. [7]

    Synaptic change in the posterior cingulate gyrus in the progression of Alzheimer's disease[J]

    Scheff S W, Price D A, Ansari M A, et al. Synaptic change in the posterior cingulate gyrus in the progression of Alzheimer's disease[J]. Journal of Alzheimer's Disease, 2015, 43(3): 1073-1090. 10 J. Zhang, X. Yu et al

    work page 2015

  7. [8]

    Longitudinal changes in fiber tract integrity in healthy aging and mild cognitive impairment: a DTI follow-up study[J]

    Teipel S J, Meindl T, Wagner M, et al. Longitudinal changes in fiber tract integrity in healthy aging and mild cognitive impairment: a DTI follow-up study[J]. Journal of Alzheimer’s Dis-ease, 2010, 22(2): 507-522

    work page 2010

  8. [9]

    Bailly M, Destrieux C, Hommet C, et al. Precuneus and Cingulate Cortex Atrophy and Hy-pometabolism in Patients with Alzheimer’s Disease and Mild Cognitive Impairment: MRI and 18F‐FDG PET Quantitative Analysis Using FreeSurfer[J]. BioMed research interna-tional, 2015, 2015(1): 583931

    work page 2015

Show all 33 references
  1. [10]

    Tau and the fractionated default mode network in atypical Alzheimer’s disease[J]

    Putcha D, Eckbo R, Katsumi Y, et al. Tau and the fractionated default mode network in atypical Alzheimer’s disease[J]. Brain Communications, 2022, 4(2): fcac055

    2022

  2. [11]

    Tau protein profiling in tauopa-thies: a human brain study[J]

    Lantero-Rodriguez J, Camporesi E, Montoliu-Gaya L, et al. Tau protein profiling in tauopa-thies: a human brain study[J]. Molecular Neurodegeneration, 2024, 19(1):

    2024

  3. [12]

    Cognitive aging outcomes are related to both tau pathol-ogy and maintenance of cingulate cortex structure[J]

    Pezzoli S, Giorgio J, Chen X, et al. Cognitive aging outcomes are related to both tau pathol-ogy and maintenance of cingulate cortex structure[J]. Alzheimer's & Dementia, 2025: e14515

    2025

  4. [13]

    The global burden of neurological disorders: translating evidence into policy[J]

    Feigin V L, Vos T, Nichols E, et al. The global burden of neurological disorders: translating evidence into policy[J]. The Lancet Neurology, 2020, 19(3): 255-265

    2020

  5. [14]

    PET 160/460 65.61 79.19 48.01 - Shao et al. (2020)

    2020

  6. [15]

    DTI 27/10 - 70.0 85.2 - Zhang et al. (2025)

    2025

  7. [16]

    rsfMRI 225/120 89.1 91.4 97.2 90/3 (B) Multi-Imaging Modality Study Zhou et al. (2021)

    2021

  8. [17]

    MRI+PET 142/82 88.25 79.74 91.58 - Shi et al. (2022)

    2022

  9. [18]

    MRI+PET 52/99 80.73 85.98 70.90 85.3 Zhang et al. (2021)

    2021

  10. [19]

    DTI+rsfMRI 116/98 92.70 - 89.90 - Zhang et al. (2023)

    2023

  11. [20]

    DTI+rsfMRI 116/93 92.30 83.22 89.99 - (C) Imaging-Non-Imaging Modality Study Janani et al. (2021)

    2021

  12. [21]

    MRI+record 132/104 84.0 - 78.5 77.0 Xue et al. (2024)

    2024

  13. [22]

    MRI+record 868/1119 59.5 55.0 64.0 41.0 Zhang et al. (2025)

    2025

  14. [23]

    MRI+text 5025/7111 - 91.5 - 89.0 Song et al. (2024)

    2024

  15. [25]

    Successful cognitive aging is associated with thicker anterior cingulate cortex and lower tau deposition compared to typical aging[J]

    Pezzoli S, Giorgio J, Martersteck A, et al. Successful cognitive aging is associated with thicker anterior cingulate cortex and lower tau deposition compared to typical aging[J]. Alz-heimer's & Dementia, 2024, 20(1): 341-355

    2024

  16. [26]

    Hypergraph based multi-task feature selection for multimodal classification of Alzheimer's disease[J]

    Shao W, Peng Y, Zu C, et al. Hypergraph based multi-task feature selection for multimodal classification of Alzheimer's disease[J]. Computerized Medical Imaging and Graphics, 2020, 80: 101663

    2020

  17. [27]

    Predicting Alzheimer’s progression in MCI: a DTI-based white matter network model[J]

    Song Q, Peng J, Shu Z, et al. Predicting Alzheimer’s progression in MCI: a DTI-based white matter network model[J]. BMC Medical Imaging, 2024, 24(1):

    2024

  18. [28]

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

    Zhang J, Lyu Y, Yu X, et al. Classification of Mild Cognitive Impairment Based on Dynamic Functional Connectivity Using Spatio-Temporal Transformer[J]. arXiv preprint arXiv:2501.16409,

    arXiv

  19. [29]

    Zhou P, Jiang S, Yu L, et al. Use of a sparse-response deep belief network and extreme learning machine to discriminate Alzheimer's disease, mild cognitive impairment, and nor-mal controls based on amyloid PET/MRI images[J]. Frontiers in Medicine, 2021, 7: 621204

    2021

  20. [30]

    ASMFS: Adaptive-similarity-based multi-modality feature se-lection for classification of Alzheimer's disease[J]

    Shi Y, Zu C, Hong M, et al. ASMFS: Adaptive-similarity-based multi-modality feature se-lection for classification of Alzheimer's disease[J]. Pattern Recognition, 2022, 126: 108566

    2022

  21. [31]

    Deep fusion of brain structure-function in mild cognitive impairment[J]

    Zhang L, Wang L, Gao J, et al. Deep fusion of brain structure-function in mild cognitive impairment[J]. Medical image analysis, 2021, 72: 102082

    2021

  22. [32]

    Multimodal deep fusion in hyperbolic space for mild cognitive impairment study[C]//International Conference on Medical Image Computing and Com-puter-Assisted Intervention

    Zhang L, Na S, Liu T, et al. Multimodal deep fusion in hyperbolic space for mild cognitive impairment study[C]//International Conference on Medical Image Computing and Com-puter-Assisted Intervention. Cham: Springer Nature Switzerland, 2023: 674-684

    2023

  23. [33]

    Multimodal deep learning models for early detection of Alzheimer’s disease stage[J]

    Venugopalan J, Tong L, Hassanzadeh H R, et al. Multimodal deep learning models for early detection of Alzheimer’s disease stage[J]. Scientific reports, 2021, 11(1):

    2021

  24. [34]

    AI-based differential diagnosis of dementia etiologies on multimodal data[J]

    Xue C, Kowshik S S, Lteif D, et al. AI-based differential diagnosis of dementia etiologies on multimodal data[J]. Nature Medicine, 2024, 30(10): 2977-2989

    2024

  25. [35]

    Brain-Adapter: Enhancing Neurological Disorder Analysis with Adapter-Tuning Multimodal Large Language Models[J]

    Zhang J, Yu X, Lyu Y, et al. Brain-Adapter: Enhancing Neurological Disorder Analysis with Adapter-Tuning Multimodal Large Language Models[J]. arXiv preprint arXiv:2501.16282,

    arXiv

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.