Personalized threshold-free tree-based localization approach for connectome-guided brain stimulation
Poster No:
121
Submission Type:
Abstract Submission
Authors:
Ruby Kong1, Aihuiping Xue1, Xiaowei Tan2, Leon Ooi1, Trevor Wei Kiat Tan1, Shan Siddiqi3, Michael Fox4, Christopher Asplund5, Bjorn Burgher6, Luca Cocchi6, Phern-Chern Tor2, B. T. Thomas Yeo1
Institutions:
1National University of Singapore, Singapore, Singapore, 2Institute of Mental Health, Singapore, Singapore, 3Harvard Medical School/Brigham and Women’s Hospital, boston, MA, 4Harvard Medical School/Brigham and Women’s Hospital, Boston, MA, 5Yale-NUS College, Singapore, Singapore, 6QIMR Berghofer Medical Research Institute, Brisbane, Queensland
First Author:
Co-Author(s):
Introduction:
Evidence suggests that individualized connectome-guided localization yields better transcranial magnetic stimulation (TMS) efficacy for treatment-resistant depression than anatomical approaches [1–4,6]. Treatment response rates improved when stimulation was delivered at dorsolateral prefrontal cortex (DLPFC) regions with stronger anti-correlation with subgenual anterior cingulate cortex (sACC) [2,4,7]. However, previous work for selecting targets require setting parameters. Here, we develop a personalized threshold-free tree-based localization approach using individual-specific functional networks.
Methods:
We use the multi-session hierarchical Bayesian model (MSHBM) to estimate reliable individual networks for each participant using resting fMRI (Fig1A)[5]. Since attention networks are known to be anticorrelated with sACC, we select the salience/ventral attention and dorsal attention network components within DLPFC. As targeting gyri close to the scalp is preferred (Fig1B), the attentional DLPFC components are further refined by only considering gyral regions based on the gyral map (Fig1C). Within the attentional DLPFC components, we select a stimulation location which is close to the scalp and anticorrelated with sACC (Fig1D). We have to set parameters for gyral map and sACC negative correlation. A tree-based approach is used to estimate a consensus location across two sets of parameters (Fig1E, 1F).
Within the attentional DLPFC components, we consider a range of gyrus thresholds (0% to 5%). For each gyrus threshold, we gradually vary the sACC correlation thresholds (100% to 5%). A x% sACC threshold represents the top x% of brain locations most anticorrelated with sACC. For given threshold, brain locations that survives are extracted, yielding one or more connected components. The centroid of each component corresponds to a tree node. For a given gyrus threshold, the centroids of the more stringent sACC correlation threshold will be the children of the centroids of the less stringent sACC correlation threshold (Fig1E).
For each tree, we select candidate targets corresponding to tree nodes with no children and tree nodes with multiple children. Among all the candidates across all trees, a final target is obtained by finding the candidate that is closest in distance on average to all other candidates (Fig1F).
To verify our localization approach, we use 2 healthy datasets. Dataset 1 had 18 local participants with two 10min runs (2 weeks apart) of resting fMRI. Dataset 2 had 32 Human Connectome Project (HCP) participants with 2 sessions of resting fMRI roughly 1 year apart. We use a single run (~15min) from each session. We compare our approach with a group-average location (Fox2012) and connectome-guided individualized locations (Cash2021).
We evaluate localizations using test-retest reliability (ratio of inter-subject target distance and intra-subject target distance) and inter-session sACC correlation (connectivity between target of one session and sACC in the other session). As test-retest reliability is ill-defined for Fox2012, the comparison is only made between Cash2021 and our approach. We perform leave-one-out cross-validation which minimizes sACC correlation for Cash2021 because Cash2021 requires setting sACC correlation threshold.
Within the attentional DLPFC components, we consider a range of gyrus thresholds (0% to 5%). For each gyrus threshold, we gradually vary the sACC correlation thresholds (100% to 5%). A x% sACC threshold represents the top x% of brain locations most anticorrelated with sACC. For given threshold, brain locations that survives are extracted, yielding one or more connected components. The centroid of each component corresponds to a tree node. For a given gyrus threshold, the centroids of the more stringent sACC correlation threshold will be the children of the centroids of the less stringent sACC correlation threshold (Fig1E).
For each tree, we select candidate targets corresponding to tree nodes with no children and tree nodes with multiple children. Among all the candidates across all trees, a final target is obtained by finding the candidate that is closest in distance on average to all other candidates (Fig1F).
To verify our localization approach, we use 2 healthy datasets. Dataset 1 had 18 local participants with two 10min runs (2 weeks apart) of resting fMRI. Dataset 2 had 32 Human Connectome Project (HCP) participants with 2 sessions of resting fMRI roughly 1 year apart. We use a single run (~15min) from each session. We compare our approach with a group-average location (Fox2012) and connectome-guided individualized locations (Cash2021).
We evaluate localizations using test-retest reliability (ratio of inter-subject target distance and intra-subject target distance) and inter-session sACC correlation (connectivity between target of one session and sACC in the other session). As test-retest reliability is ill-defined for Fox2012, the comparison is only made between Cash2021 and our approach. We perform leave-one-out cross-validation which minimizes sACC correlation for Cash2021 because Cash2021 requires setting sACC correlation threshold.
Results:
Fig2 shows the test-retest reliability and sACC correlation for both datasets. Our approach was numerically better (greater) inter/intra-subject distance than Cash2021 in both datasets, with statistical significance achieved in HCP. Our approach exhibited statistically better (more negative) sACC correlation than both Fox2012 and Cash2021.
Conclusions:
Our analyses suggest that our threshold-free tree-based approach is highly robust and led to better performance metrics across datasets compared with other approaches. This is highly desirable because this means that our approach can be easily translated to new clinics/hospitals/MRI scanners without the need to collect pilot data for tuning parameters.
Brain Stimulation:
Non-invasive Magnetic/TMS
TMS 1
Modeling and Analysis Methods:
fMRI Connectivity and Network Modeling 2
Methods Development
Keywords:
Transcranial Magnetic Stimulation (TMS)
Other - Connectivity; Individual difference; Depression
1|2Indicates the priority used for review
Provide references using author date format
2. Cash, R.F.H. et al. (2021) ‘Personalized connectivity-guided DLPFC-TMS for depression: Advancing computational feasibility, precision and reproducibility’, Human Brain Mapping, 42(13), pp. 4155–4172. Available at: https://doi.org/10.1002/hbm.25330.
3. Cole, E.J. et al. (2020) ‘Stanford Accelerated Intelligent Neuromodulation Therapy for Treatment-Resistant Depression’, American Journal of Psychiatry, 177(8), pp. 716–726. Available at: https://doi.org/10.1176/appi.ajp.2019.19070720.
4. Fox, M.D. et al. (2012) ‘Efficacy of Transcranial Magnetic Stimulation Targets for Depression Is Related to Intrinsic Functional Connectivity with the Subgenual Cingulate’, Biological Psychiatry, 72(7), pp. 595–603. Available at: https://doi.org/10.1016/j.biopsych.2012.04.028.
5. Kong, R. et al. (2019) ‘Spatial Topography of Individual-Specific Cortical Networks Predicts Human Cognition, Personality, and Emotion’, Cerebral Cortex (New York, N.Y.: 1991), 29(6), pp. 2533–2551. Available at: https://doi.org/10.1093/cercor/bhy123.
6. Siddiqi, S.H. et al. (2021) ‘Identification of Personalized Transcranial Magnetic Stimulation Targets Based on Subgenual Cingulate Connectivity: An Independent Replication’, Biological Psychiatry, 90(10), pp. e55–e56. Available at: https://doi.org/10.1016/j.biopsych.2021.02.015.
7. Weigand, A. et al. (2018) ‘Prospective Validation That Subgenual Connectivity Predicts Antidepressant Efficacy of Transcranial Magnetic Stimulation Sites’, Biological Psychiatry, 84(1), pp. 28–37. Available at: https://doi.org/10.1016/j.biopsych.2017.10.028.