Contrastive machine learning reveals Parkinson’s disease specific features
Poster No:
142
Submission Type:
Abstract Submission
Authors:
LiPing Zheng1, Cheng Zhou2, Chengjie Mao3
Institutions:
1Fudan University, Shang Hai, Shanghai Shi, 2The Second Affiliated Hospital, Zhejiang University School of Medicine, Hang Zhou, Zhe Jiang, 3The Second Affiliated Hospital of Soochow University, SuZhou, Jiang Su
First Author:
Co-Author(s):
Cheng Zhou
The Second Affiliated Hospital, Zhejiang University School of Medicine
Hang Zhou, Zhe Jiang
The Second Affiliated Hospital, Zhejiang University School of Medicine
Hang Zhou, Zhe Jiang
Introduction:
Parkinson's disease (PD) presents heterogeneous symptoms and progression due to diverse neuroanatomical changes[1]. Identifying these PD-specific alterations is vital for early stratification and personalized treatment but is complicated by factors like genetics, lifestyle, and methodological variations[2].Traditional group-level analyses often overlook patient-specific variations and nonlinear relationships. Contrastive Variational Autoencoders (CVAE) [3]address this by disentangling PD-specific features from shared neuroanatomical variations. Using structural MRI data, this study validated PD-specific features through associations with dopamine transporter (DAT) deficits, serum neurofilament light chain (NfL) levels, and clinical severity. Relevant cerebrospinal fluid (CSF) proteins and biological pathways were also identified. CVAE further enabled visualization of PD-specific neuroanatomical changes, offering insights into disease mechanisms and progression.
Methods:
This study analyzed three datasets: PPMI (646 participants), SAHZJU (581 participants), and SAHSU (71 participants for validation). Preprocessed MRI data were aligned to MNI152-2009c space and normalized to a 64×64×64 resolution. A Conditional Variational Autoencoder (CVAE) [3,4].was used to extract PD-specific features, CVAE employed two encoders to disentangle disease-specific features, unlike Variational Autoencoder's (VAE) single encoder. Synthetic brain images were generated by creating PD-specific and control brains, aligning them with ANTs software, and normalizing Jacobian maps. Dissimilarity matrices from CVAE/VAE features and clinical data were analyzed using Kendall's τ correlation and Representational Similarity Analysis (RSA). Longitudinal clinical changes were modeled using Linear Mixed Effects (LME) to assess subgroup progression. Correlations between CVAE features, clinical metrics, and cerebrospinal fluid (CSF) protein levels were evaluated. GO enrichment and STRING analyses identified biological pathways linked to PD-specific features.
Results:
PD severity measures, including putamen SBR, MDS-UPDRS-Part III, and MoCA, were primarily associated with PD-specific features rather than shared features. Similar associations were observed for fluid biomarkers, such as serum NfL, CSF α-Synuclein, T-Tau, P-Tau, and Aβ42. RSA analysis using the SAHZJU dataset confirmed these findings. PD-specific features were significantly correlated with MDS-UPDRS-Part III and MoCA slopes, highlighting their link to motor and cognitive progression in PD. Clustering PD patients based on PD-specific features using K-means identified two subgroups. LME analysis revealed that subgroup 1 had slower motor progression (MDS-UPDRS Part-III) but faster cognitive decline (MoCA) compared to subgroup 2, suggesting distinct progression patterns.We identified 1,773 CSF proteins significantly linked to PD-specific features (FDR-corrected p < 0.05), enriched in immune processes such as T cell activation, cytokine production, and neutrophil activation, as well as extracellular matrix and cytokine receptor functions. STRING analysis highlighted AKT1 as the key hub protein. Notably, 12 proteins were correlated with putamen SBR, fluid biomarkers, clinical severity, and progression.Our visualization of the PD-specific features revealed widespread neuroanatomical changes in both cortical and subcortical regions
·PD-Specific Features Modeling and its Relationship with Disease Severity and Longitudinal Clinical Progression Rates
· Association of PD-specific neuroanatomical features with CSF proteins
Conclusions:
In summary, this study used the CVAE model to disentangle PD-specific features from structural MRI data, revealing associations with well-established PD biomarkers, disease severity, and progression of both motor and cognitive symptoms. Additionally, we identified a set of CSF proteins enriched in immune function that are associated with PD-specific features. Furthermore, the CVAE model's ability to identify specific loci of neuroanatomical changes at the individual level presents an opportunity for patient stratification in precision treatments for PD.
Disorders of the Nervous System:
Neurodegenerative/ Late Life (eg. Parkinson’s, Alzheimer’s) 1
Modeling and Analysis Methods:
Other Methods 2
Keywords:
Cognition
Data analysis
Machine Learning
MRI
Other - Disease-specific variation
1|2Indicates the priority used for review
Abstract Information
By submitting your proposal, you grant permission for the Organization for Human Brain Mapping (OHBM) to distribute your work in any format, including video, audio print and electronic text through OHBM OnDemand, social media channels, the OHBM website, or other electronic publications and media.
The Open Science Special Interest Group (OSSIG) is introducing a reproducibility challenge for OHBM 2025. This new initiative aims to enhance the reproducibility of scientific results and foster collaborations between labs. Teams will consist of a “source” party and a “reproducing” party, and will be evaluated on the success of their replication, the openness of the source work, and additional deliverables. Click here for more information. Propose your OHBM abstract(s) as source work for future OHBM meetings by selecting one of the following options:
Please indicate below if your study was a "resting state" or "task-activation” study.
Healthy subjects only or patients (note that patient studies may also involve healthy subjects):
Was this research conducted in the United States?
Were any human subjects research approved by the relevant Institutional Review Board or ethics panel? NOTE: Any human subjects studies without IRB approval will be automatically rejected.
Were any animal research approved by the relevant IACUC or other animal research panel? NOTE: Any animal studies without IACUC approval will be automatically rejected.
Please indicate which methods were used in your research:
For human MRI, what field strength scanner do you use?
Which processing packages did you use for your study?
Provide references using APA citation style.
Parkinson Disease. Neurology 101, e300–e310.
[2]Mitchell, T. (2021). Emerging Neuroimaging Biomarkers across Disease Stage in Parkinson Disease: A Review. JAMA Neurol. 78,1262–1272.
[3]Aglinskas, A. (2022). Contrastive machine learning reveals the structure of neuroanatomical variation within autism. Science 376, 1070–1074.
[4]Abid, A. (2019). Contrastive Variational Autoencoder Enhances Salient Features.arXiv:1902.04601 [cs.LG]
[5] Lin, C. H. (2019). Blood NfL: A biomarker for disease severity and progression in Parkinson disease. Neurology 93, e1104–e1111
[6]Marek, K. (2018). The Parkinson’s progression markers initiative (PPMI)– establishing a PD biomarker cohort. Ann. Clin. Transl. Neurol. 5,
1460–1477.
[7]Ran, C. (2011). Genetic studies of the protein kinase AKT1 in Parkinson’s disease. Neurosci. Lett. 501, 41–44.
[8]Tian, Y. (2023). HMGB1 is a potential and challenging therapeutic target for Parkinson’s Disease. Cell Mol. Neurobiol. 43,47–58.
[9] Pan, L. (2022). Tau accelerates α-synuclein aggregation and spreading in Parkinson’s disease. Brain 145, 3454–3471.