Poster No:
1149
Submission Type:
Abstract Submission
Authors:
Ítalo Karmann Aventurato1, Lucas Scardua Silva1, Marina Koutsodontis Machado Alvim1, Rafael Batista João2, Fernando Cendes1, David Kohan Marzagão3, Clarissa Lin Yasuda1
Institutions:
1UNICAMP, Campinas, São Paulo, 2UNICAMP, Campians, São Paulo, 3King's College London, London, Greater London
First Author:
Co-Author(s):
Introduction:
Temporal Lobe Epilepsy (TLE) is a disease with significant changes in resting-state network connectivity (Zanão et al., 2021) and graph properties. However, little is known about the capacity and efficiency of graphs to differentiate patients and controls. Graph kernels represent a set of flexible and efficient methods for classification and regression problems involving graphs (K. Borgwardt et al., 2020). Here, we tested a well-known graph kernel (Weisfeiler-Lehman subtree kernel) and an in-house variation of the Shortest Path kernel (Log-Shortest Path kernel) applied to resting-state fMRI data and analyzed the efficacy/ability to classify controls and subjects with TLE.
Methods:
We used data from UNICAMP's TLE cohort (124 TLE subjects and 69 controls). BOLD sequences were obtained in a PHILIPS Achieva 3T MRI scanner with TR=2,000 ms, TE=30 ms, and a 240x240 FOV with 40 axial slices, resulting in 3 mm isotropic voxels. A total of 180 volumes (6 min. scan) were obtained for each subject. Raw BOLD data and T1 structural images were processed using the standard fMRIPrep volumetric pipeline (Esteban et al., 2019) and denoised by aggressively regressing 24 movement parameters, the top 10 aCompCor components and global signal parameters.
ROI time series were extracted using the 100 ROI version of the local-global parcellation (Schaefer et al., 2018) and used to reconstruct the partial connectivity graph using the Graphical LASSO algorithm (Friedman et al., 2008) with a penalty parameter selected using the extended Bayesian Information Criterion (Foygel & Drton, 2010). The resulting graphs were either used with weighted edges or thresholded to be used by the Weisfeiler-Lehman subtree (WL) kernel.
The WL kernel (Shervashidze et al., 2011) iteratively relabels nodes, producing label histograms at each iteration. The number of iterations (up to 5 in our implementation) is a metaparameter to be determined by cross-validation, along the edge-forming threshold (chosen among the decimal percentiles of all non-zero edges in the dataset).
Additionally, we used an in-house version of the Shortest Path kernel (K. M. Borgwardt & Kriegel, 2005) called Log-Shortest-Path (LSP). In the LSP kernel, edge weights are negative log-transformed, and path lengths are compared using positive semi-definite functions on real numbers. A scaling factor is chosen by cross-validation from 10 values geometrically distributed between 0.1 and 10. The comparing function is chosen among the Gaussian density, laplacian density, a t-distribution density with three d.f., a Brownian bridge kernel and a truncated cosine function.
Computed kernels were used to fit a Support Vector Classifier with a penalty parameter chosen from {10^-3 … 10^3} if the kernel was normalized and from {10^-6 … 10^-1}, otherwise. Performance was evaluated using a nested scheme with a 10-fold CV in the outer loop and 5 times repeated 5-fold CV in the inner loop for metaparameter selection.
Kernels were computed using three different node labelling schemes: either full labels (a label for each ROI) or the two resting-state network (RSN) labels from (Thomas Yeo et al., 2011) with either 7 or 17 labels.
Results:
Regardless of the absence of clinical information, the WL kernel obtained 64% balanced accuracy (BA) and 72% ROC-AUC with the full labelling scheme, 70% BA and 79% ROC-AUC with the 7-RSN labels, and 69% BA and 76% ROC-AUC with the 17-RSN labels.
The LSP kernel resulted in 75% BA and 82% ROC-AUC with the full labelling scheme, 71% BA and 75% ROC-AUC with the 7-RSN labels, and 64% BA and 75% ROC-AUC with the 17-RSN labels.
Conclusions:
Despite moderate accuracy, graph kernels show potential in classifying TLE using the resting-state fMRI data exclusively. In our experiment, the LSP kernel using the full labelling scheme obtained the best performance as measured by balanced accuracy (75%) and ROC-AUC (82%).
Disorders of the Nervous System:
Neurodevelopmental/ Early Life (eg. ADHD, autism) 2
Modeling and Analysis Methods:
Classification and Predictive Modeling 1
Connectivity (eg. functional, effective, structural)
fMRI Connectivity and Network Modeling
Methods Development
Keywords:
Epilepsy
Other - Resting-state fMRI, Graph Kernels
1|2Indicates the priority used for review
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.
I accept
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:
I am submitting this abstract as an original work to be reproduced. I am available to be the “source party” in an upcoming team and consent to have this work listed on the OSSIG website. I agree to be contacted by OSSIG regarding the challenge and may share data used in this abstract with another team.
Please indicate below if your study was a "resting state" or "task-activation” study.
Resting state
Healthy subjects only or patients (note that patient studies may also involve healthy subjects):
Patients
Was this research conducted in the United States?
No
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.
Yes
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.
Not applicable
Please indicate which methods were used in your research:
Functional MRI
For human MRI, what field strength scanner do you use?
3.0T
Which processing packages did you use for your study?
Other, Please list
-
fMRIPrep, NiLearn
Provide references using APA citation style.
Borgwardt, K., Ghisu, E., Llinares-López, F., O’Bray, L., & Rieck, B. (2020). Graph Kernels: State-of-the-Art and Future Challenges. In Graph Kernels: State-of-the-Art and Future Challenges. now. https://ieeexplore.ieee.org/document/9307216
Borgwardt, K. M., & Kriegel, H. P. (2005). Shortest-path kernels on graphs. Fifth IEEE International Conference on Data Mining (ICDM’05), 8 pp.-. https://doi.org/10.1109/ICDM.2005.132
Esteban, O., Markiewicz, C. J., Blair, R. W., Moodie, C. A., Isik, A. I., Erramuzpe, A., Kent, J. D., Goncalves, M., DuPre, E., Snyder, M., Oya, H., Ghosh, S. S., Wright, J., Durnez, J., Poldrack, R. A., & Gorgolewski, K. J. (2019). fMRIPrep: A robust preprocessing pipeline for functional MRI. Nature Methods, 16(1), 111–116. https://doi.org/10.1038/s41592-018-0235-4
Foygel, R., & Drton, M. (2010). Extended Bayesian Information Criteria for Gaussian Graphical Models. Advances in Neural Information Processing Systems, 23. https://proceedings.neurips.cc/paper/2010/hash/072b030ba126b2f4b2374f342be9ed44-Abstract.html
Friedman, J., Hastie, T., & Tibshirani, R. (2008). Sparse inverse covariance estimation with the graphical lasso. Biostatistics, 9(3), 432–441. https://doi.org/10.1093/biostatistics/kxm045
Schaefer, A., Kong, R., Gordon, E. M., Laumann, T. O., Zuo, X.-N., Holmes, A. J., Eickhoff, S. B., & Yeo, B. T. T. (2018). Local-Global Parcellation of the Human Cerebral Cortex from Intrinsic Functional Connectivity MRI. Cerebral Cortex, 28(9), 3095–3114. https://doi.org/10.1093/cercor/bhx179
Shervashidze, N., Schweitzer, P., van Leeuwen, E. J., Mehlhorn, K., & Borgwardt, K. M. (2011). Weisfeiler-Lehman Graph Kernels. The Journal of Machine Learning Research, 12(null), 2539–2561.
Thomas Yeo, B. T., Krienen, F. M., Sepulcre, J., Sabuncu, M. R., Lashkari, D., Hollinshead, M., Roffman, J. L., Smoller, J. W., Zöllei, L., Polimeni, J. R., Fischl, B., Liu, H., & Buckner, R. L. (2011). The organization of the human cerebral cortex estimated by intrinsic functional connectivity. Journal of Neurophysiology, 106(3), 1125–1165. https://doi.org/10.1152/jn.00338.2011
Zanão, T. A., Lopes, T. M., de Campos, B. M., Yasuda, C. L., & Cendes, F. (2021). Patterns of default mode network in temporal lobe epilepsy with and without hippocampal sclerosis. Epilepsy & Behavior: E&B, 121(Pt B), 106523. https://doi.org/10.1016/j.yebeh.2019.106523
No