An efficient connectivity-based classification model for episodic memory
Poster No:
1117
Submission Type:
Abstract Submission
Authors:
Ajay Nemani1, Yugo Boureima1, Jessica Padilla1, Christopher MacClugage1, Katherine Koenig1, Darlene Floden1, Mark Lowe1
Institutions:
1Cleveland Clinic, Cleveland, OH
First Author:
Co-Author(s):
Introduction:
Previous studies examining the relationship between episodic memory and functional connectivity measured with resting state fMRI (rsfMRI) focused on intrinsic brain networks (Suri, 2017) or specific areas such as the hippocampus (Ryan, 2010). This is because any model that explores all possible connections risks being overwhelmed by the available feature space and overfitting the data. We develop a framework that efficiently isolates a small set of connections to synthesize an accurate connectivity model for a robust memory task.
Methods:
50 cognitively healthy subjects (age = 64.2 +/- 7.0, 31 female) were selected from the Cleveland Clinic Brain Study (CCBS, 2022), a large scale, multi-year registry designed to explore risk factors and early biomarkers for neurodegenerative disease. Half the subjects scored highly (>122) on Rey's Auditory Verbal Learning Test (AVLT) (Peaker & Stewart, 1989), while the other half scored poorly (<82). Whole brain rsfMRI was performed on a 3T scanner with a 32 channel head coil (51 axial slices, SMS=3, 2.5mm isotropic voxels, TE/TR=30/1700ms, 260 volumes, 65 flip, partial Fourier 7/8). rsfMRI data was corrected for motion (Beall & Lowe, 2014), physiologically-based nuisances (Beall, 2007), and B0 distortions and registered to MNI space. The data were preprocessed with 2.5mm FWHM Gaussian spatial and 0.01-0.10 Hz temporal bandpass filters.
The group data were reduced using cohesive parcellation (Nemani, 2022), a data-driven framework designed to generate parcels with optimal exemplars. Connectivity was estimated using Pearson correlation between these exemplars. A differential classifier (Rajpoot, 2015) was developed to separate high and low memory performers. The data were randomly split in half. In the training group, a group difference connectivity matrix was generated and the 100 largest group edge differences were extracted. A support vector machine classifier was trained using these top edges as features with five-fold cross validation and tested against the validation group. This was repeated for 100 different random splits of the data. Three models were extracted from these iterations: an accuracy model based on the single iteration that performed best, a simple vote model based on the most frequent edges included across all iterations, and a weighted vote model based on the simple vote model weighted by accuracy. A final intersect model was synthesized from the overlapping edges among the three models. All four models were retrained and evaluated on the entire cohort for final evaluation.
The group data were reduced using cohesive parcellation (Nemani, 2022), a data-driven framework designed to generate parcels with optimal exemplars. Connectivity was estimated using Pearson correlation between these exemplars. A differential classifier (Rajpoot, 2015) was developed to separate high and low memory performers. The data were randomly split in half. In the training group, a group difference connectivity matrix was generated and the 100 largest group edge differences were extracted. A support vector machine classifier was trained using these top edges as features with five-fold cross validation and tested against the validation group. This was repeated for 100 different random splits of the data. Three models were extracted from these iterations: an accuracy model based on the single iteration that performed best, a simple vote model based on the most frequent edges included across all iterations, and a weighted vote model based on the simple vote model weighted by accuracy. A final intersect model was synthesized from the overlapping edges among the three models. All four models were retrained and evaluated on the entire cohort for final evaluation.
Results:
1208 parcels were generated (0.5 correlation threshold), resulting in 729632 potential connections as possible features for classification (figure 1). The accuracy, simple vote, and weighted vote models had final classification scores of 84%, 96%, and 96%, respectively (area under the ROC curve). The intersect model resulted in only 11 edges, which showed 92% classification score. Figure 2 shows this intersect model in detail, including involved parcels, underlying edges, final classifier scores, and ROC curve. The blue edges represent overlap with the overall top group edge differences over the entire cohort.
·Figure 1: Cohesive parcellation resulted in 1208 parcels (729632 edges) for classification
·Figure 2: The interect model (based on only 11 connections) showed 92% accuracy (area under the ROC curve). Blue edges overlap with the overall top group edge differences.
Conclusions:
We synthesized a highly efficient connectivity-based classifier model related to verbal memory performance. The model's features overlap with the hippocampus, entorhinal cortex, and the anterior insula, which are implicated in episodic memory function. The model suggests a possible saturation effect, with low performers showing a more linear relationship with classifier score than high performers. Further exploration of these relationships will benefit from expanding to a larger cohort as well as assessing longitudinal performance over multiple visits, both of which are available as part of the CCBS.
Learning and Memory:
Working Memory
Modeling and Analysis Methods:
Classification and Predictive Modeling 1
Connectivity (eg. functional, effective, structural) 2
Keywords:
FUNCTIONAL MRI
Memory
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?
Are you Internal Review Board (IRB) certified? Please note: Failure to have IRB, if applicable will lead to automatic rejection of abstract.
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.
Beall,E.B., Lowe,M.J. (2007). Isolating physiologic noise sources with independently determined spatial measures. Neuroimage, 37(4):1286-300. https://doi.org/10.1016/j.neuroimage.2007.07.004
Cleveland Clinic Foundation. (2022). Cleveland Clinic Brain Study. https://my.clevelandclinic.org/departments/neurological/research-innovations/brain-study
Nemani, A., & Lowe, M. J. (2022). Cohesive parcellation of the human brain using resting-state fMRI. Journal of neuroscience methods, 377, 109629. https://doi.org/10.1016/j.jneumeth.2022.109629
Peaker, A., Stewart, L.E. (1989). Rey’s Auditory Verbal Learning Test — A Review. In: Crawford, J.R., Parker, D.M. (eds), Developments in Clinical and Experimental Neuropsychology. Springer, Boston, MA. https://doi.org/10.1007/978-1-4757-9996-5_18
Rajpoot, K., Riaz, A., Majeed, W., Rajpoot, N. (2015). Functional Connectivity Alterations in Epilepsy from Resting-State Functional MRI. PloS one, 10(8), e0134944. https://doi.org/10.1371/journal.pone.0134944
Ryan, L., Lin, C. Y., Ketcham, K., Nadel, L. (2010). The role of medial temporal lobe in retrieving spatial and nonspatial relations from episodic and semantic memory. Hippocampus, 20(1), 11–18. https://doi.org/10.1002/hipo.20607
Suri, S., Topiwala, A., Filippini, N., Zsoldos, E., Mahmood, A., Sexton, C. E., Singh-Manoux, A., Kivimäki, M., Mackay, C. E., Smith, S., Ebmeier, K. P. (2017). Distinct resting-state functional connections associated with episodic and visuospatial memory in older adults. NeuroImage, 159, 122–130. https://doi.org/10.1016/j.neuroimage.2017.07.049