Targeted attacks on occipital-frontal functional connections simulates AD progression
Poster No:
1812
Submission Type:
Abstract Submission
Authors:
Avalon Campbell-Cousins1, Federica Guazzo2, Mario Parra Rodriguez3, Javier Escudero4
Institutions:
1Edinburgh University, Edinburgh, Scotland, 2University of Edinburgh, Edinburgh, Scotland, 3University of Strathclyde, Glasgow, Scotland, 4University of Edinburgh, Edinburgh, Midlothian
First Author:
Co-Author(s):
Introduction:
Alzheimer's disease (AD) is the most prominent cause of dementia and causes an immense emotional and financial burden on individuals, families, and health care services. Early stages, namely Mild Cognitive Impairment (MCI), often progresses quickly to AD [1].
Network neuroscience seeks to understand AD progression by studying brain networks constructed from neuroimaging techniques like functional Magnetic Resonance Imaging (fMRI) [2]. Damage to the brain due to AD can result in corresponding changes in functional brain connectivity [3,4]. One such model attacks connections between influential brain regions, providing a mechanism for AD progression in rs-MEG [5].
Inspired by previous research on network attacks, we present a new model to explain AD related changes in fMRI functional networks for a visual short-term memory binding task (VSTMBT). This method preferentially targets long-range connections between key brain regions involved in the task.
Network neuroscience seeks to understand AD progression by studying brain networks constructed from neuroimaging techniques like functional Magnetic Resonance Imaging (fMRI) [2]. Damage to the brain due to AD can result in corresponding changes in functional brain connectivity [3,4]. One such model attacks connections between influential brain regions, providing a mechanism for AD progression in rs-MEG [5].
Inspired by previous research on network attacks, we present a new model to explain AD related changes in fMRI functional networks for a visual short-term memory binding task (VSTMBT). This method preferentially targets long-range connections between key brain regions involved in the task.
Methods:
For this pilot study, 5 healthy participants (Age:75.4±4.7, Sex:4F) and 6 MCI that converted to AD (MCIc) after a 2-year follow up (Age:76.3±5.1, Sex:2F) were taken from a longitudinal study which acquired both diffusion and functional MRI during the VSTMBT [6], a task sensitive to memory related changes in early AD [7].
Each subject's 85x85 functional connectivity matrix is constructed using spearman correlation between each brain region's fMRI signal across repetitions of the encoding and maintenance phases of the task. fMRI data and atlas are detailed in [8].
As benchmarks, we implemented the two attack methods in [5]. 1) Random: where edges in each control network are randomly decreased by a factor of 2 until the total edge weight is the same as that of the average MCIc network. 2) Targeted: where edges are preferentially attacked when they exist between high degree nodes.
We hypothesized that long-range functional connectivity could be susceptible to damage caused by AD, much like in rs-fMRI [9], and especially those between specific brain regions key to visual short-term memory binding [10]. We devise a novel attack model to simulate the changes in early AD.
This consists of attacking the network with a probabilistic approach on the connections in the following groups: occipital-frontal (OF), occipital-parietal/temporal (OPT), frontal-parietal/temporal (FPT), and all others (AO) (method detailed in Fig. 1a). To test the effectiveness of this method, we compare the mean clustering coefficient calculated as in [5], to targeted attacks on the defined groups for a range of probabilities (Fig. 1b).
We then compare the results of the random and attack models using various graph measures (Fig. 2), for the best result in Fig. 1b. The clustering coefficient (CC) and path length (PL) are averaged over an ensemble of 50 surrogate networks created by randomly shuffling the network edge weights as in [5], with the addition of modularity (Q).
Each subject's 85x85 functional connectivity matrix is constructed using spearman correlation between each brain region's fMRI signal across repetitions of the encoding and maintenance phases of the task. fMRI data and atlas are detailed in [8].
As benchmarks, we implemented the two attack methods in [5]. 1) Random: where edges in each control network are randomly decreased by a factor of 2 until the total edge weight is the same as that of the average MCIc network. 2) Targeted: where edges are preferentially attacked when they exist between high degree nodes.
We hypothesized that long-range functional connectivity could be susceptible to damage caused by AD, much like in rs-fMRI [9], and especially those between specific brain regions key to visual short-term memory binding [10]. We devise a novel attack model to simulate the changes in early AD.
This consists of attacking the network with a probabilistic approach on the connections in the following groups: occipital-frontal (OF), occipital-parietal/temporal (OPT), frontal-parietal/temporal (FPT), and all others (AO) (method detailed in Fig. 1a). To test the effectiveness of this method, we compare the mean clustering coefficient calculated as in [5], to targeted attacks on the defined groups for a range of probabilities (Fig. 1b).
We then compare the results of the random and attack models using various graph measures (Fig. 2), for the best result in Fig. 1b. The clustering coefficient (CC) and path length (PL) are averaged over an ensemble of 50 surrogate networks created by randomly shuffling the network edge weights as in [5], with the addition of modularity (Q).
Results:
For each subject, we calculate CC for targeted attacks on the following groups: OF, OPT, FPT, and AO. We start with only attacks on AO, gradually decreasing these while increasing attacks on long-range OF connections (Fig. 1b). As we increase the attacks on OF, we see increases in CC and a closer match with our MCIc group (Fig. 1b). Next we explore the best case from Fig. 1b, comparing it to the original targeted and random attack methods in [5] for a selection of graph metrics. We find that our modified version of the targeted attack model (targeted OF in Fig. 2) clearly outperforms the original in CC and PL, and somewhat in Q.
·Fig. 1
·Fig 2.
Conclusions:
The proposed targeted attack model serves as a way to capture connectivity changes specific to the VSTMBT in early AD. Though limited by sample size, it motivates further research in how we can assess the mechanisms of AD progression. Namely, a refinement of attacks to incorporate further biological or imaging related disease mechanisms, and extensions of targeted attack methods to other imaging modalities like task-EEG.
Disorders of the Nervous System:
Neurodegenerative/ Late Life (eg. Parkinson’s, Alzheimer’s) 2
Learning and Memory:
Learning and Memory Other
Modeling and Analysis Methods:
fMRI Connectivity and Network Modeling 1
Keywords:
Computational Neuroscience
Data analysis
Degenerative Disease
FUNCTIONAL MRI
Memory
Modeling
MRI
Other - Brain Networks
1|2Indicates the priority used for review
Provide references using author date format
2) Bassett, D. S. (2018), ‘On the nature and use of models in network neuroscience’, Nature Reviews Neuroscience, vol. 19, no. 9, pp. 566-578.
3) Pereira, J. B. (2016), ‘Disrupted network topology in patients with stable and progressive mild cognitive impairment and Alzheimer's disease’, Cerebral Cortex, vol. 26, no. 8, pp. 3476-3493.
4) Sheline, Y. I. (2013), ‘Resting state functional connectivity in preclinical Alzheimer’s disease’, Biological psychiatry, vol. 74, no. 5, pp. 340-347.
5) Stam, C. J. (2009), ‘Graph theoretical analysis of magnetoencephalographic functional connectivity in Alzheimer's disease’, Brain, vol. 132, no. 1, pp. 213-224.
6) Parra, M. A. (2022), ‘Memory markers in the continuum of the Alzheimer’s clinical syndrome’, Alzheimer's research & therapy, vol. 14 no. 1, pp. 1-16.
7) Parra, M. A. (2010), ‘Visual short-term memory binding deficits in familial Alzheimer’s disease’, Brain, vol. 133, no. 9, pp. 2702-2713.
8) Fabila-Carrasco, J. S. (2023), ‘Graph-Based Permutation Patterns for the Analysis of Task-Related fMRI Signals on DTI Networks in Mild Cognitive Impairment’, arXiv preprint, arXiv:2309.13083.
9) Liu, Y. (2014), ‘Impaired long distance functional connectivity and weighted network architecture in Alzheimer's disease’, Cerebral Cortex, vol. 24, no. 6, pp. 1422-1435.
10) Parra, M. A. (2014), ‘Neural correlates of shape–color binding in visual working memory’, Neuropsychologia, vol. 52, pp. 27-36.