Network-Informed Morphometric Analysis of Alzheimer’s Disease Using Graph Signal Processing
Poster No:
1520
Submission Type:
Abstract Submission
Authors:
Yasser Alemán-Gómez1,2, Pedro Gordaliza3,1,2, Jaume Banus1,2, Meritxell Bach-Cuadra3,1,2, Patric Hagmann1,2
Institutions:
1Department of Radiology, Lausanne University Hospital (CHUV), Lausanne, Switzerland, 2Faculty of Biology and Medicine, University of Lausanne, Lausanne, Switzerland, 3CIBM Center for Biomedical Imaging, Lausanne, Switzerland
First Author:
Yasser Alemán-Gómez
Department of Radiology, Lausanne University Hospital (CHUV)|Faculty of Biology and Medicine, University of Lausanne
Lausanne, Switzerland|Lausanne, Switzerland
Department of Radiology, Lausanne University Hospital (CHUV)|Faculty of Biology and Medicine, University of Lausanne
Lausanne, Switzerland|Lausanne, Switzerland
Co-Author(s):
Pedro Gordaliza
CIBM Center for Biomedical Imaging|Department of Radiology, Lausanne University Hospital (CHUV)|Faculty of Biology and Medicine, University of Lausanne
Lausanne, Switzerland|Lausanne, Switzerland|Lausanne, Switzerland
CIBM Center for Biomedical Imaging|Department of Radiology, Lausanne University Hospital (CHUV)|Faculty of Biology and Medicine, University of Lausanne
Lausanne, Switzerland|Lausanne, Switzerland|Lausanne, Switzerland
Jaume Banus
Department of Radiology, Lausanne University Hospital (CHUV)|Faculty of Biology and Medicine, University of Lausanne
Lausanne, Switzerland|Lausanne, Switzerland
Department of Radiology, Lausanne University Hospital (CHUV)|Faculty of Biology and Medicine, University of Lausanne
Lausanne, Switzerland|Lausanne, Switzerland
Meritxell Bach-Cuadra
CIBM Center for Biomedical Imaging|Department of Radiology, Lausanne University Hospital (CHUV)|Faculty of Biology and Medicine, University of Lausanne
Lausanne, Switzerland|Lausanne, Switzerland|Lausanne, Switzerland
CIBM Center for Biomedical Imaging|Department of Radiology, Lausanne University Hospital (CHUV)|Faculty of Biology and Medicine, University of Lausanne
Lausanne, Switzerland|Lausanne, Switzerland|Lausanne, Switzerland
Patric Hagmann
Department of Radiology, Lausanne University Hospital (CHUV)|Faculty of Biology and Medicine, University of Lausanne
Lausanne, Switzerland|Lausanne, Switzerland
Department of Radiology, Lausanne University Hospital (CHUV)|Faculty of Biology and Medicine, University of Lausanne
Lausanne, Switzerland|Lausanne, Switzerland
Introduction:
Tau pathology in Alzheimer's spreads along brain networks, targeting highly connected regions 1. This network-driven spread emphasizes the need for connectivity-informed morphometric analysis, like cortical thickness (CT). Graph signal processing (GSP) is a powerful tool for integrating structure and function 2. We extend GSP to morphometry by projecting mean regional CT onto spatial harmonics derived from the Laplacian matrix of structural connectivity networks. This approach captures network-aware patterns in morphometry.
Methods:
Dataset
A total of 1,323 baseline T1-weighted images, acquired at a 3T scanner, from the Alzheimer's Disease Neuroimaging Initiative (ADNI) were analyzed. The dataset included four diagnostic groups: cognitively normal (CN: N=538, 203 males, age= 71.5±6.70 yo), early mild cognitive impairment (EMCI: N=284, 142 males, age= 71.2±7.47 yo), late MCI (LMCI: N=281, 117 males, age= 72.5±7.64 yo), and Alzheimer's disease (AD: N=220, 121 males, age= 74.5±8.27 yo).
Brain Parcellation
Each image was processed with FreeSurferv7.4.1 to create cortical surfaces and cortical thickness maps. The surfaces were subdivided using a multiscale parcellation framework3, resulting in parcellations with 68, 114, 216, and 446 cortical regions. Mean CT was calculated for each region (Figure 1A), and COMBAT4 was applied to harmonize values across acquisition sites while preserving group, age, and gender effects.
GSP Analysis
In the absence of diffusion-weighted MRI (dMRI) data, the connectivity matrix from the multiscale connectome atlas5 (scale 3) was used for graph signal processing. The graph Laplacian's eigenfunctions were used to project harmonized regional CT onto spatial harmonics (Figure 1B). Both projected and regional CT values served as independent imaging features for further analysis.
AD Classification
Random Forest6 was used to classify and identify discriminative features between diagnostic groups. The dataset was partitioned 80/20 into training and testing sets. Hyperparameter optimization was performed using 5-fold cross-validation, focusing on the number of features considered at each split (mtry: 1-25) and the ensemble size (ntrees: 501-2501, with 500-tree increments). Feature importance was assessed using both permutation-based accuracy decrease and Gini impurity measures to rank the neuroimaging predictors. The classification was performed for both regional and projected CT values.
Statistics
The Kruskal-Wallis test was applied to identify group effects on both projected and regional CT values. Pairwise comparisons were conducted using Wilcoxon tests, with Bonferroni correction for multiple comparisons.
A total of 1,323 baseline T1-weighted images, acquired at a 3T scanner, from the Alzheimer's Disease Neuroimaging Initiative (ADNI) were analyzed. The dataset included four diagnostic groups: cognitively normal (CN: N=538, 203 males, age= 71.5±6.70 yo), early mild cognitive impairment (EMCI: N=284, 142 males, age= 71.2±7.47 yo), late MCI (LMCI: N=281, 117 males, age= 72.5±7.64 yo), and Alzheimer's disease (AD: N=220, 121 males, age= 74.5±8.27 yo).
Brain Parcellation
Each image was processed with FreeSurferv7.4.1 to create cortical surfaces and cortical thickness maps. The surfaces were subdivided using a multiscale parcellation framework3, resulting in parcellations with 68, 114, 216, and 446 cortical regions. Mean CT was calculated for each region (Figure 1A), and COMBAT4 was applied to harmonize values across acquisition sites while preserving group, age, and gender effects.
GSP Analysis
In the absence of diffusion-weighted MRI (dMRI) data, the connectivity matrix from the multiscale connectome atlas5 (scale 3) was used for graph signal processing. The graph Laplacian's eigenfunctions were used to project harmonized regional CT onto spatial harmonics (Figure 1B). Both projected and regional CT values served as independent imaging features for further analysis.
AD Classification
Random Forest6 was used to classify and identify discriminative features between diagnostic groups. The dataset was partitioned 80/20 into training and testing sets. Hyperparameter optimization was performed using 5-fold cross-validation, focusing on the number of features considered at each split (mtry: 1-25) and the ensemble size (ntrees: 501-2501, with 500-tree increments). Feature importance was assessed using both permutation-based accuracy decrease and Gini impurity measures to rank the neuroimaging predictors. The classification was performed for both regional and projected CT values.
Statistics
The Kruskal-Wallis test was applied to identify group effects on both projected and regional CT values. Pairwise comparisons were conducted using Wilcoxon tests, with Bonferroni correction for multiple comparisons.
·Figure 1. Comparison of traditional ROI analysis and graph signal processing-based analysis.
Results:
Random Forest classification demonstrated similar performance in distinguishing diagnostic groups using both GSP-projected and regional CT values (Accuracy = 0.554 ±0.076 and 0.522±0.081 respectively). The highest accuracy was achieved with ntrees = 1001 and mtry = 15 in case of the projected CT values and ntrees = 2001 with mtry = 21 for regional CT values. Figure 2 shows the most important features identified by the classifier for distinguishing between groups using projected CT values (panel A) or regional CT values (panel B).
The Kruskal-Wallis test revealed significant group effects on both regional and GSP-projected CT values (Figure 2C and D). Post-hoc Wilcoxon tests identified significant differences between diagnostic groups, with the CN group exhibiting higher CT values than the AD group across key regions, including the medial temporal lobe.
The Kruskal-Wallis test revealed significant group effects on both regional and GSP-projected CT values (Figure 2C and D). Post-hoc Wilcoxon tests identified significant differences between diagnostic groups, with the CN group exhibiting higher CT values than the AD group across key regions, including the medial temporal lobe.
·Figure 2. Summary of key findings across cortical and harmonic analyses.
Conclusions:
While regional CT analysis highlights differences across distributed cortical regions, suggesting a global effect, GSP reveals connectivity-informed variations hidden within these changes. Projecting onto a subset of harmonics reduced dimensionality, providing a more comprehensive understanding of cortical alterations in Alzheimer's disease.
Lifespan Development:
Aging
Modeling and Analysis Methods:
Classification and Predictive Modeling 2
Methods Development 1
Keywords:
Aging
Morphometrics
Statistical Methods
STRUCTURAL MRI
Other - Graph Signal Processing
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.
2. Preti, M.G., Van De Ville, D. Decoupling of brain function from structure reveals regional behavioral specialization in humans. Nat Commun 10, 4747 (2019).
3. Cammoun L, Gigandet X, Meskaldji D, Thiran JP, Sporns O, Do KQ, Maeder P, Meuli R, Hagmann P. Mapping the human connectome at multiple scales with diffusion spectrum MRI. J Neurosci Methods. 2012 Jan 30;203(2):386-97
4. Fortin JP, Cullen N, Sheline YI, Taylor WD, Aselcioglu I, Cook PA, Adams P, Cooper C, Fava M, McGrath PJ, McInnis M, Phillips ML, Trivedi MH, Weissman MM, Shinohara RT. Harmonization of cortical thickness measurements across scanners and sites. Neuroimage. 2018 Feb 15;167:104-120.
5. Alemán-Gómez, Y., Griffa, A., Houde, JC. et al. A multi-scale probabilistic atlas of the human connectome. Sci Data 9, 516 (2022).
6. Ho, T.K., Random decision forests. In Proceedings of 3rd international conference on document analysis and recognition. pp. 278–282 (1995).