Explaining Unsupervised Disease Staging in Huntington's Disease: Insights into Model Representations and Clusters

Hind Zantout; Lubna Mahmoud Abu Zohair

arxiv: 2606.07135 · v2 · pith:BTJUFWSBnew · submitted 2026-06-05 · 💻 cs.LG

Explaining Unsupervised Disease Staging in Huntington's Disease: Insights into Model Representations and Clusters

Lubna Mahmoud Abu Zohair , Hind Zantout This is my paper

Pith reviewed 2026-06-27 22:19 UTC · model grok-4.3

classification 💻 cs.LG

keywords Huntington's diseaseunsupervised learningexplainabilitySHAPdisease stagingembeddingsEnroll-HD dataset

0 comments

The pith

Unsupervised embeddings for Huntington's disease staging align with clinical motor and functional scores.

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

The paper extends an unsupervised machine learning framework for staging Huntington's disease by adding explainability analysis to the learned representations and clusters. Projections of the embeddings show clusters that track progressive worsening in established clinical measures. Saliency maps and SHAP values from a surrogate classifier identify the clinical features driving the embeddings and stage assignments. The analysis finds that the stages range from early cognitive-motor impairment to severe functional dependency in patterns that match known disease progression.

Core claim

The explainability analysis indicates that the learned embeddings capture clinically meaningful disease structure, aligning with established motor and functional severity scores and exhibiting progressive deterioration across clusters. SHAP reveals a stratification of disease stages, ranging from early cognitive-motor impairment to severe functional dependency, consistent with known clinical progression patterns, while also highlighting intra-stage variability.

What carries the argument

Surrogate classifier with SHAP applied to the unsupervised embeddings to quantify feature contributions to cluster assignments.

If this is right

The discovered clusters exhibit progressive deterioration consistent with clinical measures.
Motor and functional scores are the primary drivers of the learned embeddings over time.
SHAP identifies transitions between stages from early impairment to severe dependency.
The embeddings exhibit intra-stage variability within the overall progression pattern.

Where Pith is reading between the lines

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

The same explainability approach could be applied to unsupervised staging models for other progressive diseases to check for similar clinical alignment.
If the surrogate approximation holds, the identified features could be prioritized in future data collection for HD monitoring.
The stratification suggests the model could support finer-grained tracking of stage transitions in longitudinal studies.

Load-bearing premise

The surrogate classifier produces SHAP values that accurately reflect the original unsupervised model's internal feature contributions.

What would settle it

Re-training a different surrogate model on the same embeddings yields substantially different SHAP rankings for the same clusters, or new patient data shows clusters that fail to align with motor and functional severity scores.

Figures

Figures reproduced from arXiv: 2606.07135 by Hind Zantout, Lubna Mahmoud Abu Zohair.

**Figure 1.** Figure 1: UMAP projection of the learned embeddings, coloured by (A) the corresponding cluster number, and clinical severity scores for (B) motor and (C) functional measures. adl (0.77), indicating marked loss of independence. Motor impairment features such as dystlue (0.76), gait (0.73), and dystlle (0.71) were also highly ranked, alongside bulbar involvement reflected by tongue (0.57). This combination of featur… view at source ↗

**Figure 2.** Figure 2: Saliency map for clinical features importance across age. Rows represent input features, while columns represent age/time points. Feature importance values are normalized, and hierarchical clustering is applied to group features with similar temporal saliency patterns. tive changes, indicating their growing influence in defining more advanced disease stages. Finally, the transition from Cluster 2 to Clust… view at source ↗

**Figure 3.** Figure 3: Cluster-specific clinical measurement importance based on SHAP analysis. For each cluster, feature contributions were quantified by averaging the absolute SHAP values of samples assigned to that cluster, using the SHAP values corresponding to the predicted cluster class. The top features, selected based on their overall importance, are shown. Each line represents a feature, and values indicate its relative… view at source ↗

**Figure 4.** Figure 4: SHAP-based feature transition analysis across disease stages. Heatmap showing ∆SHAP values (difference in mean absolute SHAP importance) for all clinical features across consecutive cluster transitions (0→1, 1→2, 2→3). Positive values (red) indicate increasing feature influence, while negative values (blue) indicate decreasing influence across stages [PITH_FULL_IMAGE:figures/full_fig_p012_4.png] view at source ↗

read the original abstract

Huntington's disease (HD) is a progressive neurodegenerative disorder that affects motor, cognitive, and behavioral functions, where accurate characterization of disease progression remains essential to improve patient outcome and quality of life. Unsupervised machine learning (ML) approaches have demonstrated the ability to uncover disease progression trajectories and meaningful latent stages from longitudinal data; however, their limited interpretability restricts clinical trust and translation. We extend a previously proposed ML-based disease staging framework by applying an explainability analysis to the extracted feature representations and discovered disease stages. Applied to the Enroll-HD dataset, we first project the learned representations into a lower-dimensional space to intuitively assess whether the resulting clusters align with the progression of established clinical measures. We then use saliency maps to identify the clinical features that most strongly contribute to the learned embeddings over time. Finally, we train a surrogate classifier and apply SHAP to quantify feature importance for cluster assignments and to analyze which clinical variables drive transitions between disease stages. The explainability analysis indicates that the learned embeddings capture clinically meaningful disease structure, aligning with established motor and functional severity scores and exhibiting progressive deterioration across clusters. Within this analysis, SHAP reveals a stratification of disease stages, ranging from early cognitive-motor impairment to severe functional dependency, consistent with known clinical progression patterns, while also highlighting intra-stage variability.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

Applies standard saliency and SHAP to a prior unsupervised HD staging pipeline on Enroll-HD but supplies no numbers or surrogate checks to support the alignment claims.

read the letter

The paper takes an existing unsupervised staging method, runs it on the Enroll-HD dataset, projects the embeddings, and layers on saliency maps plus SHAP from a surrogate classifier to explain the clusters. The main observation is that the clusters track motor and functional scores and that SHAP points to expected progression from cognitive-motor issues to functional dependency.

What it does is straightforward: it shows one way to inspect whether the learned representations line up with clinical measures and which features appear to drive the stage assignments. That kind of check can be helpful when people want to see if unsupervised clusters make clinical sense.

The gaps are clear. The abstract states alignment and stratification but gives no correlation values, statistical tests, confidence intervals, or error bars. The surrogate step is unexamined; nothing reports how accurately it recovers the original cluster assignments or whether it preserves the geometry of the embeddings. The stress-test note is right on this point. Without that check, the reported feature importances for stage transitions rest on an unverified approximation. The work also stays inside the same clinical variables used to train the embeddings, so the explanations risk echoing patterns already present in the input.

This is for people already focused on ML staging for Huntington's or similar diseases who are looking for another application example. It does not introduce new techniques or deliver quantitative evidence that would move the field. I would not bring it to a reading group or cite it. It does not look ready for peer review without the missing validation numbers and surrogate fidelity checks.

Referee Report

3 major / 1 minor

Summary. The manuscript extends a prior unsupervised ML disease staging framework for Huntington's disease by applying explainability methods to embeddings learned from the Enroll-HD longitudinal dataset. It projects embeddings to assess cluster alignment with clinical progression, uses saliency maps to identify feature contributions to embeddings over time, and trains a surrogate classifier to enable SHAP analysis of feature importance for cluster assignments and stage transitions. The central claim is that the embeddings capture clinically meaningful structure (aligning with motor/functional scores and showing progressive deterioration) and that SHAP reveals stratification from early cognitive-motor impairment to severe functional dependency consistent with known patterns.

Significance. If the surrogate-based SHAP attributions and alignment claims hold after verification, the work would strengthen interpretability of unsupervised disease staging models, potentially increasing clinical trust and translation for progressive disorders like HD by linking latent representations to established clinical measures.

major comments (3)

[Abstract / Methods] Abstract and Methods (surrogate classifier description): The central claim that SHAP 'reveals a stratification of disease stages' and 'quantif[ies] feature importance for cluster assignments' rests on a surrogate classifier trained to predict the unsupervised clusters, yet no fidelity metrics (accuracy, calibration, or agreement with original embedding geometry) are reported; any systematic mismatch would directly distort the reported feature importances and intra-stage variability.
[Abstract] Abstract (results claims): The assertion that embeddings 'align with established motor and functional severity scores' and exhibit 'progressive deterioration across clusters' is presented without quantitative support such as correlation coefficients, statistical tests, or error bars, which is load-bearing for the 'clinically meaningful disease structure' conclusion.
[Abstract / Discussion] Abstract / Discussion: The analysis learns representations from the same clinical features later used to evaluate alignment with motor/functional scores and to interpret SHAP values, creating a circularity risk that is not addressed with controls or sensitivity checks; this undermines the independence of the 'consistent with known clinical progression patterns' claim.

minor comments (1)

[Abstract] The abstract would be clearer if it specified the number of discovered clusters/stages, the exact clinical scores (e.g., UHDRS motor, TFC), and the dimensionality reduction technique used for visualization.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for their constructive feedback, which highlights important aspects of methodological transparency and claim support. We address each major comment below with point-by-point responses. Revisions will be incorporated to strengthen the manuscript where indicated.

read point-by-point responses

Referee: [Abstract / Methods] Abstract and Methods (surrogate classifier description): The central claim that SHAP 'reveals a stratification of disease stages' and 'quantif[ies] feature importance for cluster assignments' rests on a surrogate classifier trained to predict the unsupervised clusters, yet no fidelity metrics (accuracy, calibration, or agreement with original embedding geometry) are reported; any systematic mismatch would directly distort the reported feature importances and intra-stage variability.

Authors: We agree that fidelity metrics are critical to validate the surrogate and ensure SHAP attributions are trustworthy. The full manuscript describes the surrogate training but does not explicitly tabulate fidelity metrics in the provided sections. In the revision, we will add a dedicated subsection reporting accuracy, calibration plots, and agreement (e.g., adjusted Rand index) between surrogate predictions and original cluster assignments, confirming high fidelity to support the reported SHAP results. revision: yes
Referee: [Abstract] Abstract (results claims): The assertion that embeddings 'align with established motor and functional severity scores' and exhibit 'progressive deterioration across clusters' is presented without quantitative support such as correlation coefficients, statistical tests, or error bars, which is load-bearing for the 'clinically meaningful disease structure' conclusion.

Authors: We acknowledge that the abstract would benefit from explicit quantitative anchors. The full manuscript contains correlation analyses and statistical tests in the Results, but these are not summarized numerically in the abstract. We will revise the abstract to include key quantitative results (e.g., Spearman correlation coefficients with motor/functional scores and associated p-values) while maintaining brevity. revision: yes
Referee: [Abstract / Discussion] Abstract / Discussion: The analysis learns representations from the same clinical features later used to evaluate alignment with motor/functional scores and to interpret SHAP values, creating a circularity risk that is not addressed with controls or sensitivity checks; this undermines the independence of the 'consistent with known clinical progression patterns' claim.

Authors: This concern about circularity is substantive. The unsupervised model discovers latent structure without supervision from the clinical scores, and post-hoc alignment with specific motor/functional measures is a standard validation approach. However, to directly address potential dependence, the revised manuscript will include sensitivity analyses (e.g., ablation of motor/functional features from the input set followed by re-evaluation of cluster alignment) and report whether progressive patterns remain consistent. revision: yes

Circularity Check

0 steps flagged

No significant circularity; external clinical benchmarks keep derivation independent

full rationale

The paper applies unsupervised clustering to Enroll-HD longitudinal data, projects embeddings for visual alignment against established motor/functional severity scores, generates saliency maps on the learned representations, and trains a surrogate classifier solely to enable SHAP interpretation of the resulting cluster labels. These steps compare discovered structure to independent clinical measures and known progression patterns rather than deriving claims tautologically from fitted inputs or self-citations. No equations reduce a claimed prediction to its own training targets by construction, no uniqueness theorems are imported from prior author work, and no ansatz is smuggled via citation. The chain remains self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review supplies no explicit free parameters, axioms, or invented entities; the central claim rests on the unstated assumption that the unsupervised model and surrogate faithfully represent clinical reality.

pith-pipeline@v0.9.1-grok · 5766 in / 1070 out tokens · 19714 ms · 2026-06-27T22:19:07.789369+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

22 extracted references · 15 canonical work pages · 3 internal anchors

[1]

org/10.1016/S0166-2236(00)01568-X

Jakel,R.J.,Maragos,W.F.:NeuronalcelldeathinHuntington’sdisease:apotential role for dopamine.Trends in Neurosciences23(6), 239–245 (2000).https://doi. org/10.1016/S0166-2236(00)01568-X

work page doi:10.1016/s0166-2236(00)01568-x 2000
[2]

Abu Zohair, L.M., Zantout, H., Vallejo, M., Uddin, M.A., Mahmoud, A.: Arti- ficial intelligence and machine learning for the diagnosis of Huntington disease: a narrative review.Journal of Medical Artificial Intelligence9(2025).https: //jmai.amegroups.org/article/view/10176

2025
[3]

Abu Zohair, L.M., Zantout, H., Vallejo, M., Uddin, M.A.: Unsupervised machine learning using cerebrospinal fluid proteomics for understanding Parkinson’s disease progression.Proceedings of the AAAI Symposium Series6(1),72–74(2025).https: //doi.org/10.1609/aaaiss.v6i1.36033

work page doi:10.1609/aaaiss.v6i1.36033 2025
[4]

Ross, C.A., Tabrizi, S.J.: Huntington’s disease: from molecular pathogenesis to clinical treatment.The Lancet Neurology10(1), 83–98 (2011).https://doi.org/ 10.1016/S1474-4422(10)70245-3

work page doi:10.1016/s1474-4422(10)70245-3 2011
[5]

Shoulson, I., Fahn, S.: Huntington disease: clinical care and evaluation.Neurology 29(1), 1–3 (1979).https://doi.org/10.1212/WNL.29.1.1

work page doi:10.1212/wnl.29.1.1 1979
[6]

Tabrizi,S.J.etal.:AbiologicalclassificationofHuntington’sdisease:theIntegrated Staging System.The Lancet Neurology21(7), 632–644 (2022).https://doi.org/ 10.1016/S1474-4422(22)00120-X

work page doi:10.1016/s1474-4422(22)00120-x 2022
[7]

et al.: Rating scales for cognition in Huntington’s disease: critique and recommendations.Movement Disorders33(2) (2018).https://doi.org/10

Mestre, T.A. et al.: Rating scales for cognition in Huntington’s disease: critique and recommendations.Movement Disorders33(2) (2018).https://doi.org/10. 1002/mds.27227

2018
[8]

https://doi.org/10.1002/mdc3.12618 14 L

Winder, J.Y., Roos, R.A.C., Burgunder, J.-M., Marinus, J., Reilmann, R.: Inter- rater reliability of the Unified Huntington’s Disease Rating Scale – Total Motor Score certification.Movement Disorders Clinical Practice5(3), 290–295 (2018). https://doi.org/10.1002/mdc3.12618 14 L. Abu Zohair and H. Zantout

work page doi:10.1002/mdc3.12618 2018
[9]

et al.: Refining the language of Huntington’s disease progression with HD-ISS.Journal of Huntington’s Disease13(2), 115–118 (2024).https://doi

Sampaio, C. et al.: Refining the language of Huntington’s disease progression with HD-ISS.Journal of Huntington’s Disease13(2), 115–118 (2024).https://doi. org/10.3233/JHD-240043

work page doi:10.3233/jhd-240043 2024
[10]

et al.: Classification of manifest Huntington disease using vowel distor- tion measures

Romana, A. et al.: Classification of manifest Huntington disease using vowel distor- tion measures. In:Interspeech(2020), pp. 4966–4970.https://doi.org/10.21437/ Interspeech.2020-2724

2020
[11]

0000000000200430

Paulsen, J.S., Burks, D.K.: Revised diagnostic criteria for Huntington disease.Neu- rology Clinical Practice15(2), e200430 (2025).https://doi.org/10.1212/CPJ. 0000000000200430

work page doi:10.1212/cpj 2025
[12]

Abu Zohair, L.M., Vallejo, M., Uddin, M.A., Woodward, J.R., Zantout, H.: A Machine Learning-Based Framework for Discovering Huntington’s Disease Stages: Integrating Graph Representation Learning and Clustering to Uncover Pro- gression Dynamics in Longitudinal Enroll-HD Dataset. 2026. arXiv:2606.06196. https://arxiv.org/abs/2606.06196

work page internal anchor Pith review Pith/arXiv arXiv 2026
[13]

Lubna Mahmoud Abuzohair, Hind Zantout, Md Azher Uddin et al. A Novel Dy- namic Graph Architecture for Staging Parkinson’s Disease Progression Using Cere- brospinal Fluids Longitudinal Profiles, 08 March 2026, PREPRINT (Version 1) availableatResearchSquareResearch Square(2026).https://doi.org/10.21203/ rs.3.rs-9034342/v1

2026
[14]

Ren, Z. et al.: Regularized feature reconstruction for spatio-temporal saliency de- tection.IEEE Transactions on Image Processing22(8), 3120–3132 (2013).https: //doi.org/10.1109/TIP.2013.2259837

work page doi:10.1109/tip.2013.2259837 2013
[15]

UMAP: Uniform Manifold Approximation and Projection for Dimension Reduction

McInnes, L., Healy, J., Melville, J.: UMAP: Uniform Manifold Approximation and Projection for Dimension Reduction. arXiv preprint arXiv:1802.03426 (2020). https://arxiv.org/abs/1802.03426

work page internal anchor Pith review Pith/arXiv arXiv 2020
[16]

Enroll-HD: Enroll-HD Data Dictionary. (2025).https:// enroll-hd.org/for-researchers/data-support-documentation/ #documentation/doc-details2/68b60a0a59371b02d0740572/kn-asset/ 103-206-137-68b609da96de8802d6a30de6/enrollhd_datadictionary_ 20250807.xlsx

2025
[17]

CHDI Foundation: PDS7 Overview: Dataset overview. (2025).https: //enroll-hd.org/for-researchers/data-support-documentation/ #documentation/doc-details2/68ac787cf252ac044f5a0833/kn-asset/ 103-206-137-6900e9ff885d080310120b70/enrollhd_pds7overview_20251028. pdf

2025
[18]

CHDI Foundation: Enroll-HD acknowledgement list.https:// enroll-hd.org/for-researchers/data-support-documentation/ #documentation/doc-details2/68ac76f628f56202b991fc57/kn-asset/ 103-206-137-68ac76cc1c6fff02b835133a/enrollhd_acknowledgementlist_ 20250821.pdf
[19]

Deep Inside Convolutional Networks: Visualising Image Classification Models and Saliency Maps

Simonyan, K., Vedaldi, A., Zisserman, A.: Deep inside convolutional networks: visualising saliency maps. arXiv preprint arXiv:1312.6034 (2014).https://arxiv. org/abs/1312.6034

work page internal anchor Pith review Pith/arXiv arXiv 2014
[20]

et al.: From clustering to cluster explanations via neural networks

Kauffmann, J. et al.: From clustering to cluster explanations via neural networks. IEEE Transactions on Neural Networks and Learning Systems35(2), 1926–1940 (2024).https://doi.org/10.1109/TNNLS.2022.3185901

work page doi:10.1109/tnnls.2022.3185901 1926
[21]

Ross, C.A., Aylward, E.H., Wild, E.J., Langbehn, D.R., Long, J.D., Warner, J.H., Scahill, R.I., Leavitt, B.R., Stout, J.C., Paulsen, J.S., Reilmann, R., Unschuld, P.G., Wexler, A., Margolis, R.L., Tabrizi, S.J.: Huntington disease: natural history, biomarkers and prospects for therapeutics.Nature Reviews Neurology10(4), 204– 216 (2014).https://doi.org/10....

work page doi:10.1038/nrneurol.2014.24 2014
[22]

The Huntington’s Disease Collaborative Research Group: Huntington disease.Na- ture Reviews Disease Primers1, 15052 (2015).https://doi.org/10.1038/nrdp. 2015.52

work page doi:10.1038/nrdp 2015

[1] [1]

org/10.1016/S0166-2236(00)01568-X

Jakel,R.J.,Maragos,W.F.:NeuronalcelldeathinHuntington’sdisease:apotential role for dopamine.Trends in Neurosciences23(6), 239–245 (2000).https://doi. org/10.1016/S0166-2236(00)01568-X

work page doi:10.1016/s0166-2236(00)01568-x 2000

[2] [2]

Abu Zohair, L.M., Zantout, H., Vallejo, M., Uddin, M.A., Mahmoud, A.: Arti- ficial intelligence and machine learning for the diagnosis of Huntington disease: a narrative review.Journal of Medical Artificial Intelligence9(2025).https: //jmai.amegroups.org/article/view/10176

2025

[3] [3]

Abu Zohair, L.M., Zantout, H., Vallejo, M., Uddin, M.A.: Unsupervised machine learning using cerebrospinal fluid proteomics for understanding Parkinson’s disease progression.Proceedings of the AAAI Symposium Series6(1),72–74(2025).https: //doi.org/10.1609/aaaiss.v6i1.36033

work page doi:10.1609/aaaiss.v6i1.36033 2025

[4] [4]

Ross, C.A., Tabrizi, S.J.: Huntington’s disease: from molecular pathogenesis to clinical treatment.The Lancet Neurology10(1), 83–98 (2011).https://doi.org/ 10.1016/S1474-4422(10)70245-3

work page doi:10.1016/s1474-4422(10)70245-3 2011

[5] [5]

Shoulson, I., Fahn, S.: Huntington disease: clinical care and evaluation.Neurology 29(1), 1–3 (1979).https://doi.org/10.1212/WNL.29.1.1

work page doi:10.1212/wnl.29.1.1 1979

[6] [6]

Tabrizi,S.J.etal.:AbiologicalclassificationofHuntington’sdisease:theIntegrated Staging System.The Lancet Neurology21(7), 632–644 (2022).https://doi.org/ 10.1016/S1474-4422(22)00120-X

work page doi:10.1016/s1474-4422(22)00120-x 2022

[7] [7]

et al.: Rating scales for cognition in Huntington’s disease: critique and recommendations.Movement Disorders33(2) (2018).https://doi.org/10

Mestre, T.A. et al.: Rating scales for cognition in Huntington’s disease: critique and recommendations.Movement Disorders33(2) (2018).https://doi.org/10. 1002/mds.27227

2018

[8] [8]

https://doi.org/10.1002/mdc3.12618 14 L

Winder, J.Y., Roos, R.A.C., Burgunder, J.-M., Marinus, J., Reilmann, R.: Inter- rater reliability of the Unified Huntington’s Disease Rating Scale – Total Motor Score certification.Movement Disorders Clinical Practice5(3), 290–295 (2018). https://doi.org/10.1002/mdc3.12618 14 L. Abu Zohair and H. Zantout

work page doi:10.1002/mdc3.12618 2018

[9] [9]

et al.: Refining the language of Huntington’s disease progression with HD-ISS.Journal of Huntington’s Disease13(2), 115–118 (2024).https://doi

Sampaio, C. et al.: Refining the language of Huntington’s disease progression with HD-ISS.Journal of Huntington’s Disease13(2), 115–118 (2024).https://doi. org/10.3233/JHD-240043

work page doi:10.3233/jhd-240043 2024

[10] [10]

et al.: Classification of manifest Huntington disease using vowel distor- tion measures

Romana, A. et al.: Classification of manifest Huntington disease using vowel distor- tion measures. In:Interspeech(2020), pp. 4966–4970.https://doi.org/10.21437/ Interspeech.2020-2724

2020

[11] [11]

0000000000200430

Paulsen, J.S., Burks, D.K.: Revised diagnostic criteria for Huntington disease.Neu- rology Clinical Practice15(2), e200430 (2025).https://doi.org/10.1212/CPJ. 0000000000200430

work page doi:10.1212/cpj 2025

[12] [12]

Abu Zohair, L.M., Vallejo, M., Uddin, M.A., Woodward, J.R., Zantout, H.: A Machine Learning-Based Framework for Discovering Huntington’s Disease Stages: Integrating Graph Representation Learning and Clustering to Uncover Pro- gression Dynamics in Longitudinal Enroll-HD Dataset. 2026. arXiv:2606.06196. https://arxiv.org/abs/2606.06196

work page internal anchor Pith review Pith/arXiv arXiv 2026

[13] [13]

Lubna Mahmoud Abuzohair, Hind Zantout, Md Azher Uddin et al. A Novel Dy- namic Graph Architecture for Staging Parkinson’s Disease Progression Using Cere- brospinal Fluids Longitudinal Profiles, 08 March 2026, PREPRINT (Version 1) availableatResearchSquareResearch Square(2026).https://doi.org/10.21203/ rs.3.rs-9034342/v1

2026

[14] [14]

Ren, Z. et al.: Regularized feature reconstruction for spatio-temporal saliency de- tection.IEEE Transactions on Image Processing22(8), 3120–3132 (2013).https: //doi.org/10.1109/TIP.2013.2259837

work page doi:10.1109/tip.2013.2259837 2013

[15] [15]

UMAP: Uniform Manifold Approximation and Projection for Dimension Reduction

McInnes, L., Healy, J., Melville, J.: UMAP: Uniform Manifold Approximation and Projection for Dimension Reduction. arXiv preprint arXiv:1802.03426 (2020). https://arxiv.org/abs/1802.03426

work page internal anchor Pith review Pith/arXiv arXiv 2020

[16] [16]

Enroll-HD: Enroll-HD Data Dictionary. (2025).https:// enroll-hd.org/for-researchers/data-support-documentation/ #documentation/doc-details2/68b60a0a59371b02d0740572/kn-asset/ 103-206-137-68b609da96de8802d6a30de6/enrollhd_datadictionary_ 20250807.xlsx

2025

[17] [17]

CHDI Foundation: PDS7 Overview: Dataset overview. (2025).https: //enroll-hd.org/for-researchers/data-support-documentation/ #documentation/doc-details2/68ac787cf252ac044f5a0833/kn-asset/ 103-206-137-6900e9ff885d080310120b70/enrollhd_pds7overview_20251028. pdf

2025

[18] [18]

CHDI Foundation: Enroll-HD acknowledgement list.https:// enroll-hd.org/for-researchers/data-support-documentation/ #documentation/doc-details2/68ac76f628f56202b991fc57/kn-asset/ 103-206-137-68ac76cc1c6fff02b835133a/enrollhd_acknowledgementlist_ 20250821.pdf

[19] [19]

Deep Inside Convolutional Networks: Visualising Image Classification Models and Saliency Maps

Simonyan, K., Vedaldi, A., Zisserman, A.: Deep inside convolutional networks: visualising saliency maps. arXiv preprint arXiv:1312.6034 (2014).https://arxiv. org/abs/1312.6034

work page internal anchor Pith review Pith/arXiv arXiv 2014

[20] [20]

et al.: From clustering to cluster explanations via neural networks

Kauffmann, J. et al.: From clustering to cluster explanations via neural networks. IEEE Transactions on Neural Networks and Learning Systems35(2), 1926–1940 (2024).https://doi.org/10.1109/TNNLS.2022.3185901

work page doi:10.1109/tnnls.2022.3185901 1926

[21] [21]

Ross, C.A., Aylward, E.H., Wild, E.J., Langbehn, D.R., Long, J.D., Warner, J.H., Scahill, R.I., Leavitt, B.R., Stout, J.C., Paulsen, J.S., Reilmann, R., Unschuld, P.G., Wexler, A., Margolis, R.L., Tabrizi, S.J.: Huntington disease: natural history, biomarkers and prospects for therapeutics.Nature Reviews Neurology10(4), 204– 216 (2014).https://doi.org/10....

work page doi:10.1038/nrneurol.2014.24 2014

[22] [22]

The Huntington’s Disease Collaborative Research Group: Huntington disease.Na- ture Reviews Disease Primers1, 15052 (2015).https://doi.org/10.1038/nrdp. 2015.52

work page doi:10.1038/nrdp 2015