Pith. sign in

REVIEW 4 major objections 5 minor 45 references

Single Domain Generalization for Alzheimer's Detection from 3D MRIs with Pseudo-Morphological Augmentations and Contrastive Learning

T0 review · 4 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read A 3D U-Net trained only on NACC, with learnable pseudo-erosion and pseudo-dilation augmentations plus supervised contrastive learning, beats generic single-domain-generalization baselines on ADNI and AIBL.

desk verdict The combination of pseudo-morphological augmentations and contrastive learning is a plausible new SDG pipeline for AD detection, but the ablation that supposedly isolates the morphology modules is confounded, so the central mechanistic claim is currently unsupported. read the letter →

arxiv 2505.22465 v2 pith:W6J4GDV2 submitted 2025-05-28 cs.CV

Zobia Batool , Huseyin Ozkan , Erchan Aptoula This is my paper
classification cs.CV
keywords singledomaingeneralizationAlzheimer'sdiseasedetection3DMRIpseudo-morphologicalaugmentationserosionanddilationsupervisedcontrastivelearningclassimbalanceshift
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

The paper asks whether a model trained on a single MRI dataset can detect Alzheimer's disease reliably on scans from unseen sites with different scanners and protocols. Its proposal is to generate class-specific training images with learnable pseudo-morphological operators, namely erosion to simulate the tissue loss of Alzheimer's and dilation to simulate healthy brain volume, and to pair them with a weighted supervised contrastive loss. Trained only on NACC, the resulting 3D U-Net achieves the highest macro-F1 among compared methods on both ADNI (0.424) and AIBL (0.456), suggesting that anatomy-aware augmentation plus contrastive learning is a workable recipe for single-domain generalization under class imbalance.

What carries the argument

The load-bearing mechanism is the pair of learnable pseudo-morphological modules built from 3D convolutions followed by channel-wise max for dilation and the negation of a channel-wise min for erosion, with binary foreground masks so background zeros do not drive the operations. These approximate the classical morphological operators of mathematical morphology without a full lattice foundation; erosion is intended to simulate brain atrophy and dilation to simulate healthy expansion. Around them, a weighted supervised contrastive loss with class weights pulls same-class projections together and pushes other classes apart, while CutMix3D is applied to uniform MCI batches for extra intra-class diversity. Together the modules generate anatomically motivated class-specific augmentations during training, and the encoder learns representations that remain discriminative on unseen target distributions.

What would settle it

Compare the spatial pattern of changes induced by the pseudo-erosion module on healthy control MRIs against longitudinal atrophy maps of patients who later convert to Alzheimer's; if the simulated changes do not overlap known atrophy regions such as the hippocampus and medial temporal lobe significantly more than random intensity perturbations do, the anatomical mechanism is not supported.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is that differentiable, class-aware morphological augmentation transfers to unseen imaging domains better than generic augmentation or feature-alignment baselines. Erosion and dilation are treated as disease models: applying learnable erosion to Alzheimer's samples intensifies atrophy, applying learnable dilation to healthy controls expands brain tissue, and affine augmentations plus 3D CutMix handle borderline mild cognitive impairment cases. These class-specific views are fed through a shared 3D U-Net encoder with a supervised contrastive loss, and the whole pipeline is trained end-to-end on NACC alone. On ADNI the method reaches 50.91% accuracy and macro-F1 0.424; on AIBL it reaches macro-F1 0.456, with gains over the strongest baseline on both balanced metrics. The paper interprets the results as evidence that morphological priors preserve disease-relevant structure while increasing intra-class diversity, which is exactly what an unseen-domain detector needs.

Load-bearing premise

The method's gains depend on the assumption that the learned erosion and dilation modules produce anatomically plausible class-specific changes, namely genuine atrophy and healthy brain expansion, rather than generic intensity distortions.

Editorial extensions

If this is right

  • A model trained only on NACC can be deployed directly on ADNI and AIBL scans without any target-domain data or labels, since the single-domain-generalization protocol uses no target examples.
  • Morphology-based augmentations improve balanced metrics such as macro-F1 more than generic MixUp, RSC, CCSDG, or EFDM under class imbalance.
  • Removing the morphological modules drops macro-F1 by 6.5 points on ADNI and 11.8 points on AIBL in the ablations, indicating that the erosion and dilation views carry much of the generalization gain.
  • Grad-CAM maps produced with the morphological modules concentrate on hippocampus and medial temporal lobe regions, indicating more disease-relevant localization.
  • The added cost is modest in parameters (21.2M versus 19.6M for the baseline) but heavier in FLOPs and memory, and the paper treats this as an acceptable trade-off for the F1 gains.

Reading between the lines

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

  • A natural next test, not run in the paper, is whether the same class-aware erosion and dilation recipe transfers to other neurodegenerative diseases with known atrophy patterns, such as frontotemporal dementia or Parkinson's disease.
  • If the learned morphological operators are truly anatomy-based, their per-voxel effect on healthy scans should overlap known atrophy maps; quantifying that overlap would provide an independent check of the mechanism.
  • The method's AIBL accuracy is lower than EFDM's, so a multi-objective evaluation that also optimizes accuracy could change the ranking; the paper's case rests on balanced macro-F1.
  • With a larger multi-site training cohort, the contrastive and morphological components could be tested separately at scale, for instance by ablating only erosion or only dilation.
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. This manuscript proposes a single-domain generalization (SDG) method for Alzheimer's disease detection from 3D MRIs. A 3D U-Net encoder is trained on NACC only; class-specific pseudo-dilation and pseudo-erosion modules generate augmented NC/AD samples, MCI samples receive affine and CutMix3D augmentation, and a supervised contrastive loss is optimized jointly with weighted cross-entropy. The method is evaluated on ADNI and AIBL and compared with MixUp, RSC, CCSDG, EFDM, and a no-SDG baseline; the authors report the highest macro-F1 on both target datasets and provide an ablation study, computational cost comparison, and Grad-CAM visualizations.

Significance. If the claims hold, the paper offers a well-motivated combination of morphological priors and supervised contrastive learning for single-domain generalization in neuroimaging, with macro-F1 as an appropriate primary metric under class imbalance. The use of three public datasets and a standard source-only evaluation protocol is a strength, and the paper attempts to ablate each component. However, the central empirical claim is currently supported by a mislabeled ablation row and by single-run numbers without error bars or significance tests, and the paper does not release code, so the significance of the contribution will only be clear after the experimental accounting is corrected.

major comments (4)
  1. [Table 4, Sec. 4.4] The first row of Table 4 ('No pseudo-morphology') is numerically identical to the Baseline row in Tables 2 and 3 (ADNI ACC 38.04, F1 0.359, SEN 0.359, SPE 0.679; AIBL ACC 38.50, F1 0.338, SEN 0.392, SPE 0.699). Since the baseline described in Sec. 4.2 has no SDG components at all (no CutMix3D, no contrastive loss), this row cannot represent removing only the pseudo-morphological modules while keeping the other components fixed. Therefore the statement in Sec. 4.4 that removing morphological modules causes F1 drops of 6.5 points on ADNI and 11.8 points on AIBL is unsupported. This is the central attribution of the paper, and the row must either be recomputed as a true single-component ablation or the text must be corrected.
  2. [Sec. 4.3, Tables 2 and 3] All results appear to be single-run numbers with no error bars, confidence intervals, or significance tests. The claimed improvement over the strongest baseline is small on ADNI (F1 0.424 vs. 0.407, a 1.7-point difference), and the claim that the method 'consistently achieved superior generalization performance' is also contradicted by Table 3, where EFDM reaches 69.94% accuracy on AIBL versus 62.27% for the proposed method. The authors should report means and standard deviations over at least three seeds for the main comparisons, and should qualify the superiority claim by specifying the metric (macro-F1, sensitivity, specificity) rather than accuracy.
  3. [Algorithm 1, Sec. 3.3] Algorithm 1 as printed contains unreachable code and is inconsistent with Sec. 3.3. When a minibatch has all labels equal, the pseudocode returns LCE after applying CutMix (for MCI) or immediately (for non-MCI), so the subsequent construction of the augmented set A, the computation of LSCL, and the combined loss Ltotal are never executed for such batches. When the batch has mixed labels, the homogeneous-batch branch is skipped. This makes it unclear when the supervised contrastive loss is actually applied, despite Eq. (6) assuming 2N original-plus-augmentation samples. The algorithm should be rewritten so that every code path and the exact loss composition are unambiguous.
  4. [Table 4, Sec. 4.4] The 'No supervised contrastive loss' ablation row reports AIBL accuracy of 73.28%, which is 11.01 points higher than the full method's 62.27%, while the macro-F1 drops from 0.456 to 0.305 and sensitivity from 0.452 to 0.343. The text only discusses the F1 and sensitivity drops and does not address this large accuracy swing. The authors should explain whether the full method intentionally trades accuracy for balanced performance, or whether this result indicates an unstable training run, and should provide a class-wise breakdown for this setting.
minor comments (5)
  1. [References] References [16] and [17] appear to describe the same paper: both are titled 'Learning with domain-knowledge for generalizable prediction of Alzheimer's disease from multi-site structural MRI' and both are listed as MICCAI 2023, Vancouver, pages 452–461. One of them should be removed or replaced with a distinct work.
  2. [Fig. 5] The caption of Figure 5 is malformed: the line 'xa xa xa (c) ˜xa' is not a complete caption and does not explain what is shown in each panel.
  3. [Eqs. (2) and (4)] Equations (2) and (4) are taken from the morphological module design in [23]; the text should state this explicitly in the contribution summary so the novelty is clearly the SDG application and the class-specific augmentation strategy, not the differentiable morphology operator itself.
  4. [Table 5] The 'Model size (MB)' of 494.56 MB is inconsistent with the reported 21.2M parameters (which would be about 85 MB as fp32). Please define whether this column measures GPU memory footprint during inference rather than parameter storage, and report both consistently.
  5. [Sec. 3.2.1] The foreground mask is written as 'M = 1 x̸=0'; this should use proper indicator notation, e.g., M = 1[x ≠ 0], and the mask application order should be stated precisely for both the dilation and erosion modules.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: the paper's claims are empirical comparisons, and the only self-citation is not load-bearing.

full rationale

The paper does not claim a first-principles derivation whose output is built into its inputs. Its two novel-looking components are (i) differentiable pseudo-dilation/pseudo-erosion modules explicitly adopted from external prior work [23], used as class-conditioned training-time augmentations, and (ii) a standard supervised contrastive loss [36] combined with weighted cross-entropy. Equations (2) and (4) are presented as borrowed module definitions, not as evidence for generalization, so there is no self-definitional step. No fitted parameter is relabeled as a prediction: accuracy, F1, sensitivity, and specificity are measured on held-out ADNI and AIBL after training on NACC, so the central experimental claims are independent of the method's own outputs. The only self-citation, [33], appears in Related Work as a description of a prior distance-transform mixup strategy and is not used to justify the proposed architecture, augmentations, or loss; it is therefore not load-bearing. One non-circularity concern should be flagged for correctness rather than for the circularity score: in Table 4, the 'No pseudo-morphology' row is numerically identical to the 'Baseline [35]' row in Tables 2 and 3 (ADNI ACC 38.04 / F1 0.359 / SEN 0.359 / SPE 0.679; AIBL ACC 38.50 / F1 0.338 / SEN 0.392 / SPE 0.699), and Sec. 4.4 interprets this row as removing only the morphological modules while retaining CutMix and the contrastive loss. If those components were retained, the ablation row should differ from the no-SDG baseline. This is an internal-validity or table-consistency problem in the ablation attribution, not a case where a claimed prediction reduces by construction to its inputs, so it does not change the circularity finding.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The central claim rests on the anatomical validity of pseudo-morphological augmentations and on the sufficiency of NACC as a source domain; both are assumed, not independently verified. No new physical entities are introduced.

free parameters (3)
  • temperature tau
    Temperature in the supervised contrastive loss (Eq. 6); the value is not reported and is chosen by hand.
  • contrastive loss weight lambda
    Balances cross-entropy and contrastive losses in L_total; the value is not reported.
  • pseudo-morphology kernel size k = {3, 5}
    Random k in {3,5} is chosen to simulate anatomical variation at different scales; it is not optimized or empirically justified.
assumptions (4)
  • domain assumption Grayscale erosion and dilation mimic AD atrophy and healthy brain expansion
    Central motivation in Sec. 3.2, with no expert or quantitative validation that outputs resemble true disease states.
  • ad hoc to paper The foreground mask M = 1_{x != 0} confines augmentations to anatomically relevant tissue
    Introduced to preserve boundaries; no evidence of anatomical validity is provided.
  • domain assumption NACC is a representative single source domain for SDG evaluation
    The SDG protocol in Sec. 4.2 assumes an 80/20 NACC split is sufficient and ADNI/AIBL are unseen; no domain-shift magnitude analysis is given.
  • domain assumption The pretrained Models Genesis 3D U-Net is an appropriate backbone
    The encoder is chosen from [35] without comparison to other backbones.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Single Domain Generalization for Alzheimer's Detection from 3D MRIs with Pseudo-Morphological Augmentations and Contrastive Learning." pith.science (2026). https://pith.science/paper/W6J4GDV2

@misc{pith2026250522465,
  author       = {Pith},
  title        = {Pith review of: Single Domain Generalization for Alzheimer's Detection from 3D MRIs with Pseudo-Morphological Augmentations and Contrastive Learning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/W6J4GDV2}},
  note         = {Machine review of arXiv:2505.22465}
}
read the original abstract

Although Alzheimer's disease detection via MRIs has advanced significantly thanks to contemporary deep learning models, challenges such as class imbalance, protocol variations, and limited dataset diversity often hinder their generalization capacity. To address this issue, this article focuses on the single domain generalization setting, where given the data of one domain, a model is designed and developed with maximal performance w.r.t. an unseen domain of distinct distribution. Since brain morphology is known to play a crucial role in Alzheimer's diagnosis, we propose the use of learnable pseudo-morphological modules aimed at producing shape-aware, anatomically meaningful class-specific augmentations in combination with a supervised contrastive learning module to extract robust class-specific representations. Experiments conducted across three datasets show improved performance and generalization capacity, especially under class imbalance and imaging protocol variations. The source code will be made available upon acceptance at https://github.com/zobia111/SDG-Alzheimer.

Figures

Figures reproduced from arXiv: 2505.22465 by the authors.

Figure 1
Figure 1. Overview of the proposed training pipeline. Class-specific augmentations (erosion, dilation, or [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Encoder architecture of 3D U-Net. 3.2 Class-Specific Augmentations To address distributional shifts between training and testing domains, class-specific learnable morphological 3D augmentation is incorporated. While standard augmentation techniques that apply identical transforms may improve robustness, they often miss disease-specific changes, potentially blurring class differences. To prevent this, the proposed me… view at source ↗
Figure 3
Figure 3. Visualization of a pseudo-dilation result. (a) Original NC image from the dataset. (b) Resulting [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Visualization of a pseudo-erosion result. (a) Original AD image from the dataset. (b) Resulting [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 5
Figure 5. Figure 5: CutMix pipeline: (a) Source image xa, (b) Source image xb, (c) Region-wise patches are extracted and swapped to generate mixed image ˜xa in 3D space. 3.3 Weighted Supervised Contrastive Learning To encourage domain-invariant and class-discriminative representations, th…
Figure 6
Figure 6. Figure 6: Grad-CAM visualizations on MRI scans from ADNI and AIBL datasets. (a) Input images for NC, [PITH_FULL_IMAGE:figures/full_fig_p010_6.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

45 extracted references · 43 canonical work pages

  1. [16]

    Y. Zhou, Y. Li, F. Zhou, Y. Liu, and L. Tu. Learning with domain-knowledge for generalizable prediction of Alzheimer’s disease from multi-site structural MRI. In Proceedings of the International Conference on Medical Image Computing and Computer-Assisted Intervention , pages 452–461, Vancouver, Canada, 2023

    work page 2023

  2. [17]

    Learning with domain-knowledge for generalizable prediction of Alzheimer’s disease from multi-site structural MRI

    Yanjie Zhou, Youhao Li, Feng Zhou, Yong Liu, and Liyun Tu. Learning with domain-knowledge for generalizable prediction of Alzheimer’s disease from multi-site structural MRI. In Proceedings of the International Conference on Medical Image Computing and Computer-Assisted Intervention , pages 452–461, Vancouver, Canada, 2023

    work page 2023

  3. [1]

    Disentangle first, then distill: a unified framework for missing modality imputation and Alzheimer’s disease diagnosis

    Yuanyuan Chen, Yongsheng Pan, Yong Xia, and Yixuan Yuan. Disentangle first, then distill: a unified framework for missing modality imputation and Alzheimer’s disease diagnosis. IEEE Transactions on Medical Imaging, 42(12):3566–3578, 2023

    work page 2023

  4. [2]

    Generalizing to unseen domains: A survey on domain generalization

    Jindong Wang, Cuiling Lan, Chang Liu, Yidong Ouyang, Tao Qin, Wang Lu, Yiqiang Chen, Wenjun Zeng, and S Yu Philip. Generalizing to unseen domains: A survey on domain generalization. IEEE transactions on knowledge and data engineering , 35(8):8052–8072, 2022

    work page 2022

  5. [3]

    Multi-template meta-information regularized network for Alzheimer’s disease diagnosis using structural MRI

    Kangfu Han, Gang Li, Zhiwen Fang, and Feng Yang. Multi-template meta-information regularized network for Alzheimer’s disease diagnosis using structural MRI. IEEE Transactions on Medical Imaging , 43(5):1664–1676, 2023

    work page 2023

  6. [4]

    A multimodal approach integrating convolutional and recurrent neural networks for Alzheimer’s disease temporal progression prediction

    Durga Supriya Hl, Swetha Mary Thomas, et al. A multimodal approach integrating convolutional and recurrent neural networks for Alzheimer’s disease temporal progression prediction. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 5207–5215, 2024

    work page 2024

  7. [5]

    M3t: three-dimensional medical image classifier using multi-plane and multi-slice transformer

    Jinseong Jang and Dosik Hwang. M3t: three-dimensional medical image classifier using multi-plane and multi-slice transformer. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 20718–20729, 2022

    work page 2022

  8. [6]

    Multi-attention- based global 3d resnet for Alzheimer’s disease diagnosis

    Yaozu Li, Yueheng Zhang, Jinfeng Wu, Xiaoshuang Zhang, Lili Han, and Xinchun Cui. Multi-attention- based global 3d resnet for Alzheimer’s disease diagnosis. In 2024 International Joint Conference on Neural Networks, pages 1–8, 2024

    work page 2024

Show all 45 references
  1. [7]

    3d global fourier network for Alzheimer’s disease diagnosis using structural MRI

    Shengjie Zhang, Xiang Chen, Bohan Ren, Haibo Yang, Ziqi Yu, Xiao-Yong Zhang, and Yuan Zhou. 3d global fourier network for Alzheimer’s disease diagnosis using structural MRI. In Proceedings of the International Conference on Medical Image Computing and Computer-Assisted Interve...

    2022

  2. [8]

    An effective deep learning-based automatic prediction and classification of Alzheimer’s disease using EGELU-SZN technique

    B Sathyabhama and M Kannan. An effective deep learning-based automatic prediction and classification of Alzheimer’s disease using EGELU-SZN technique. Neural Computing and Applications , pages 1–18, 2025

    2025

  3. [9]

    Vitad: Leveraging modified vision transformer for Alzheimer’s disease multi-stage classification from brain MRI scans

    Md Ashif Mahmud Joy, Shamima Nasrin, Ayesha Siddiqua, and Dewan Md Farid. Vitad: Leveraging modified vision transformer for Alzheimer’s disease multi-stage classification from brain MRI scans. Brain Research, 1847:149302, 2025

    2025

  4. [10]

    Multimodal surface-based transformer model for early diagnosis of Alzheimer’s disease

    Quan Anh Duong, Sy Dat Tran, and Jin Kyu Gahm. Multimodal surface-based transformer model for early diagnosis of Alzheimer’s disease. Scientific Reports, 15(1):5787, 2025

    2025

  5. [11]

    Training vit with limited data for Alzheimer’s disease classification: An empirical study

    Kassymzhomart Kunanbayev, Vyacheslav Shen, and Dae-Shik Kim. Training vit with limited data for Alzheimer’s disease classification: An empirical study. In Proceedings of the International Conference on Medical Image Computing and Computer-Assisted Intervention , pages 334–343,...

    2024

  6. [12]

    Dense convolution-based attention network for Alzheimer’s disease classification

    Yingtong Gan, Quan Lan, ChenXi Huang, Weichao Su, and Zhiyuan Huang. Dense convolution-based attention network for Alzheimer’s disease classification. Scientific Reports, 15(1):5693, 2025

    2025

  7. [13]

    Anatomy-aware gating network for explainable Alzheimer’s disease diagnosis

    Hongchao Jiang and Chunyan Miao. Anatomy-aware gating network for explainable Alzheimer’s disease diagnosis. In Proceedings of the International Conference on Medical Image Computing and Computer- Assisted Intervention, pages 90–100, Marrakesh, Marocco, 2024. 11

    2024

  8. [14]

    Lteif, S

    D. Lteif, S. Sreerama, S. A. Bargal, B. A. Plummer, R. Au, and V. B. Kolachalama. Disease-driven domain generalization for neuroimaging-based assessment of Alzheimer’s disease. Human Brain Mapping , 45(8):e26707, Jun. 2024

    2024

  9. [15]

    Prototype-guided multi-scale domain adaptation for Alzheimer’s disease detection

    Hongshun Cai, Qiongmin Zhang, and Ying Long. Prototype-guided multi-scale domain adaptation for Alzheimer’s disease detection. Computers in Biology and Medicine , 154:106570, 2023

    2023

  10. [18]

    Mathematical morphology

    Jean Serra. Mathematical morphology. In Encyclopedia of mathematical geosciences , pages 820–835. Springer, 2023

    2023

  11. [19]

    Histopathological image analysis: A review

    Metin N Gurcan, Laura E Boucheron, Ali Can, Anant Madabhushi, Nasir M Rajpoot, and Bulent Yener. Histopathological image analysis: A review. IEEE reviews in biomedical engineering , 2:147–171, 2009

    2009

  12. [20]

    Deformable medical image registration: A survey

    Aristeidis Sotiras, Christos Davatzikos, and Nikos Paragios. Deformable medical image registration: A survey. IEEE Transactions on Medical Imaging , 32(7):1153–1190, 2013

    2013

  13. [21]

    Medical image fusion: A survey of the state of the art

    Alex Pappachen James and Belur V Dasarathy. Medical image fusion: A survey of the state of the art. Information fusion , 19:4–19, 2014

    2014

  14. [22]

    Multivariate mathematical morphology applied to colour image analysis

    Erchan Aptoula and S´ ebastien Lef` evre. Multivariate mathematical morphology applied to colour image analysis. Multivariate image processing: methods and applications , pages 303–337, 2009

    2009

  15. [23]

    Learning deep morphological networks with neural architecture search

    Yufei Hu, Nacim Belkhir, Jesus Angulo, Angela Yao, and Gianni Franchi. Learning deep morphological networks with neural architecture search. Pattern Recognition, 131:108893, 2022

    2022

  16. [24]

    Multi-scale morphology-aided deep medical image segmentation

    Susmita Ghosh and Swagatam Das. Multi-scale morphology-aided deep medical image segmentation. Engineering Applications of Artificial Intelligence , 137:109047, 2024

    2024

  17. [25]

    Differentiable soft morphological filters for medical image segmentation

    Lisa Guzzi, Maria A Zuluaga, Fabien Lareyre, Gilles Di Lorenzo, S´ ebastien Goffart, Andrea Chierici, Juliette Raffort, and Herv´ e Delingette. Differentiable soft morphological filters for medical image segmentation. In Proceedings of the International Conference on Medical I...

    2024

  18. [26]

    Hybrid morphological-convolutional neural networks for computer-aided diagnosis

    Martha Rebeca Canales-Fiscal and Jos´ e Gerardo Tamez-Pe˜ na. Hybrid morphological-convolutional neural networks for computer-aided diagnosis. Frontiers in Artificial Intelligence , 6:1253183, 2023

    2023

  19. [27]

    Topological properties of individual gray matter morphological networks in identifying the preclinical stages of Alzheimer’s disease: a preliminary study

    Hongyuan Ding, Zhihao Wang, Yin Tang, Tong Wang, Ming Qi, Weiqiang Dou, Long Qian, Yaxin Gao, Qian Zhong, Xi Yang, et al. Topological properties of individual gray matter morphological networks in identifying the preclinical stages of Alzheimer’s disease: a preliminary study. ...

    2023

  20. [28]

    Interpretation of brain morphology in association to Alzheimer’s disease dementia classification using graph convolutional networks on triangulated meshes

    Emanuel A Azcona, Pierre Besson, Yunan Wu, Arjun Punjabi, Adam Martersteck, Amil Dravid, Todd B Parrish, S Kathleen Bandt, and Aggelos K Katsaggelos. Interpretation of brain morphology in association to Alzheimer’s disease dementia classification using graph convolutional netw...

    2020

  21. [29]

    Adapt: Alzheimer diagnosis through adaptive profiling transformers

    Yifeng Wang, Ke Chen, and Haohan Wang. Adapt: Alzheimer diagnosis through adaptive profiling transformers. arXiv preprint arXiv:2401.06349 , 2024. 12

    2024 arXiv

  22. [30]

    Embracing the disharmony in medical imaging: A simple and effective framework for domain adaptation

    Rongguang Wang, Pratik Chaudhari, and Christos Davatzikos. Embracing the disharmony in medical imaging: A simple and effective framework for domain adaptation. Medical image analysis , 76:102309, 2022

    2022

  23. [31]

    Multi-site MRI harmonization via attention-guided deep domain adaptation for brain disorder identification

    Hao Guan, Yunbi Liu, Erkun Yang, Pew-Thian Yap, Dinggang Shen, and Mingxia Liu. Multi-site MRI harmonization via attention-guided deep domain adaptation for brain disorder identification. Medical image analysis , 71:102076, 2021

    2021

  24. [32]

    Towards better interpretable and generalizable ad detection using collective artificial intelligence

    Huy-Dung Nguyen, Micha¨ el Cl´ ement, Boris Mansencal, and Pierrick Coup´ e. Towards better interpretable and generalizable ad detection using collective artificial intelligence. Computerized Medical Imaging and Graphics, 104:102171, 2023

    2023

  25. [33]

    Distance transform guided mixup for Alzheimer’s detection

    Zobia Batool, Huseyin Ozkan, and Erchan Aptoula. Distance transform guided mixup for Alzheimer’s detection. In IEEE Conference on Signal Processing and Communications Applications, Istanbul, T¨ urkiye,

  26. [34]

    Cutmix: Regularization strategy to train strong classifiers with localizable features

    Sangdoo Yun, Dongyoon Han, Seong Joon Oh, Sanghyuk Chun, Junsuk Choe, and Youngjoon Yoo. Cutmix: Regularization strategy to train strong classifiers with localizable features. In Proceedings of the IEEE/CVF International Conference on Computer Vision , pages 6023–6032, 2019

    2019

  27. [35]

    Z. Zhou, V. Sodha, M. M. R. Siddiquee, R. Feng, N. Tajbakhsh, M. B. Gotway, and J. Liang. Models Genesis: Generic autodidactic models for 3D medical image analysis. In Proceedings of the International Conference on Medical Image Computing and Computer-Assisted Intervention, pa...

    2019

  28. [36]

    Supervised contrastive learning

    Prannay Khosla, Piotr Teterwak, Chen Wang, Aaron Sarna, Yonglong Tian, Phillip Isola, Aaron Maschinot, Ce Liu, and Dilip Krishnan. Supervised contrastive learning. Advances in neural information processing systems, 33:18661–18673, 2020

    2020

  29. [37]

    D. L. Beekly, E. M. Ramos, G. van Belle, W. Deitrich, A. D. Clark, M. E. Jacka, and W. A. Kukull. The national Alzheimer’s coordinating center (NACC) database: an Alzheimer disease database. Alzheimer Disease & Associated Disorders , 18(4):270–277, 2004

    2004

  30. [38]

    R. C. Petersen, P. S. Aisen, L. A. Beckett, M. C. Donohue, A. C. Gamst, D. J. Harvey, C. R. Jack, W. J. Jagust, L. M. Shaw, A. W. Toga, and J. Q. Trojanowski. Alzheimer’s disease neuroimaging initiative (ADNI) clinical characterization. Neurology, 74(3):201–209, 2010

    2010

  31. [39]

    K. A. Ellis, A. I. Bush, D. Darby, D. De Fazio, J. Foster, P. Hudson, N. T. Lautenschlager, N. Lenzo, R. N. Martins, R. Maruff, P. Masters, and the AIBL Research Group. The Australian imaging, biomarkers and lifestyle (AIBL) study of aging: methodology and baseline characteris...

    2009

  32. [40]

    Multimodal deep learning for Alzheimer’s disease dementia assessment

    Shangran Qiu, Matthew I Miller, Prajakta S Joshi, Joyce C Lee, Chonghua Xue, Yunruo Ni, Yuwei Wang, Ileana De Anda-Duran, Phillip H Hwang, Justin A Cramer, et al. Multimodal deep learning for Alzheimer’s disease dementia assessment. Nature communications, 13(1):3404, 2022

    2022

  33. [41]

    Learning to learn single domain generalization

    Fengchun Qiao, Long Zhao, and Xi Peng. Learning to learn single domain generalization. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 12556–12565, 2020

    2020

  34. [42]

    H. Zhang. Mixup: Beyond empirical risk minimization. arXiv preprint, 2017

    2017

  35. [43]

    Huang, H

    Z. Huang, H. Wang, E. P. Xing, and D. Huang. Self-challenging improves cross-domain generalization. In Proceedings of the European Conference on Computer Vision , pages 124—-140, Glasgow, UK, 2020

    2020

  36. [44]

    Exact feature distribution matching for arbitrary style transfer and domain generalization

    Yabin Zhang, Minghan Li, Ruihuang Li, Kui Jia, and Lei Zhang. Exact feature distribution matching for arbitrary style transfer and domain generalization. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 8035–8045, 2022

    2022

  37. [45]

    Devil is in channels: Contrastive single domain generalization for medical image segmentation

    Shishuai Hu, Zehui Liao, and Yong Xia. Devil is in channels: Contrastive single domain generalization for medical image segmentation. In Proceedings of the International Conference on Medical Image Computing and Computer-Assisted Intervention , pages 14–23, Vancouver, Canada, 2023. 13

    2023

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.