Comparing Four Brain Age Algorithms in Predicting Cognition in Bipolar Disorder and Healthy Persons
Poster No:
582
Submission Type:
Abstract Submission
Authors:
Hui Xin Ng1, Lisa Eyler1
Institutions:
1University of California San Diego, La Jolla, CA
First Author:
Co-Author:
Introduction:
Our study compares four Brain Age algorithms in their ability to predict level of cognitive performance among individuals with bipolar disorder (BD) versus healthy comparisons (HC). We investigate whether higher brain-predicted age difference (brain-PAD) is linked to poorer cognition in BD, given BD's neuropathological similarities to brain aging (Baecker et al., 2021). We expected the BD group to exhibit a stronger negative correlation between brain-PAD and cognition. We hypothesized that algorithms trained on more granular data would better capture group differences in the strength of the brain-PAD-cognition relationship and exhibit more robust brain-PAD-cognition associations, irrespective of group.
Methods:
Dataset: The dataset includes 38 HC and 33 individuals with BD, who met the criteria for BD I and were euthymic. Groups were comparable in terms of age, sex, and education. Cognitive performance was assessed using the Delis-Kaplan Executive Function System (DKEFS) test. Cognitive scores included: Trails Visual Scanning Omission Errors (Misses) and Commission Errors (False Alarm), Trails Completion Times (Number Sequencing, Letter-Number, Motor Speed), Color Word Interference (Color Naming, Word Reading, Inhibition, Inhibition/Switching). Image acquisition: General Electric Signa EXCITE 3.0 Tesla whole-body imaging system in a one session with a variety of T1-weighted and T2-weighted structural and functional sequences.
Algorithm Types: 3 algorithms were selected due to their large training sets, public code and diversity of training data granularity and algorithm type: PHOTON-BA (Freesurfer parcellated T1-weighted scans; Han et al., 2021), BrainageR (voxel-based T1-weighted scans; Cole et al., 2018), MIDI (axial T2-weighted and axial diffusion-weighted scans; Wood et al., 2022). UCSD Multimodal, a locally developed ridge regression model, included multiple modalities like T1-weighted MRI, DTI, CBF, task fMRI, and rsMRI.
Statistical Analysis: Principal component analysis (PCA) on cognitive scores resulting in a first component (PC1) accounted for 44.5% of variance. We demographically-corrected each algorithm's brain-PAD by adjusting for sex, age, and their interaction; residuals from respective linear models were used in subsequent analyses to test our hypotheses. Residual brain-PAD (rbrain-PAD), group, and their interaction were used as predictors of PC1. Interaction term effect sizes were compared. A model with only rbrain-PAD was then fit given the prior non-significance of group and interaction. We compared effect sizes across simplified models.
Algorithm Types: 3 algorithms were selected due to their large training sets, public code and diversity of training data granularity and algorithm type: PHOTON-BA (Freesurfer parcellated T1-weighted scans; Han et al., 2021), BrainageR (voxel-based T1-weighted scans; Cole et al., 2018), MIDI (axial T2-weighted and axial diffusion-weighted scans; Wood et al., 2022). UCSD Multimodal, a locally developed ridge regression model, included multiple modalities like T1-weighted MRI, DTI, CBF, task fMRI, and rsMRI.
Statistical Analysis: Principal component analysis (PCA) on cognitive scores resulting in a first component (PC1) accounted for 44.5% of variance. We demographically-corrected each algorithm's brain-PAD by adjusting for sex, age, and their interaction; residuals from respective linear models were used in subsequent analyses to test our hypotheses. Residual brain-PAD (rbrain-PAD), group, and their interaction were used as predictors of PC1. Interaction term effect sizes were compared. A model with only rbrain-PAD was then fit given the prior non-significance of group and interaction. We compared effect sizes across simplified models.
Results:
In the model with group, rbrain-PAD and their interaction, we found no significant associations of rbrain-PAD or group with PC1 across all algorithms (all p's >0.05). There was no significant interaction of group and rbrain-PAD across the algorithms, but effect sizes of the interaction varied, with the strongest association seen for BrainageR (b= -0.06, t= -1.98, p= 0.05, ηp2= 0.056); MIDI (b= -0.09, t= -1.73, p= 0.08, ηp2=0.043); UCSD Multimodal (b = -0.04, t= -0.99, p= 0.32, ηp2= 0.015); PHOTON-BA (b= 0.00015, t= 0.004, p= 0.99, ηp2= 0.00). With BD and HC groups combined as a follow-up, we found that rbrain-PAD was significantly related to cognition for one algorithm (BrainageR; b=-0.04, t= -2.62, p < 0.05, ηp2= 0.091) but not others (MIDI b = -0.04, t= -1.84, p= 0.07, ηp2= 0.047;UCSD Multimodal b= -0.03, t= -1.69, p= 0.09, ηp2= 0.040; PHOTON-BA b = -0.03, t= 0.08, ηp2= 0.043).
Conclusions:
Contrary to our hypothesis, we did not see significantly stronger negative brain-PAD-cognition link in BD compared to HC across all algorithms, but ones trained on raw data like BrainageR and MIDI show potential in detecting a overall brain-PAD-cognition relationship and differential strength of that relationship in BD. BrainageR performed the best in both aspects, showing its potential for revealing deeper relationships between brain aging and cognition among those with serious mental illness such as BD.
Disorders of the Nervous System:
Psychiatric (eg. Depression, Anxiety, Schizophrenia) 1
Higher Cognitive Functions:
Executive Function, Cognitive Control and Decision Making
Lifespan Development:
Aging
Modeling and Analysis Methods:
Multivariate Approaches 2
Neuroinformatics and Data Sharing:
Informatics Other
Keywords:
Aging
Cognition
Computational Neuroscience
DISORDERS
1|2Indicates the priority used for review
Provide references using author date format
Cole J.H., Ritchie S.J., Bastin M.E., et al (2018) Brain age predicts mortality. Mol Psychiatry 23:1385– 1392
Han, L. K. M., Dinga, R., Hahn, T., Ching, C. R. K., Eyler, L. T., Aftanas, L., Aghajani, M., Aleman, A., Baune, B. T., Berger, K., Brak, I., Filho, G. B., Carballedo, A., Connolly, C. G., Couvy-Duchesne, B., Cullen, K. R., Dannlowski, U., Davey, C. G., Dima, D., … Schmaal, L. (2021). Brain aging in major depressive disorder: Results from the ENIGMA major depressive disorder working group. Molecular Psychiatry, 26(9), Article 9. https://doi.org/10.1038/s41380-020-0754-0
Wood, D. A., Kafiabadi, S., Busaidi, A. A., Guilhem, E., Montvila, A., Lynch, J., Townend, M., Agarwal, S., Mazumder, A., Barker, G. J., Ourselin, S., Cole, J. H., & Booth, T. C. (2022). Accurate brain‐age models for routine clinical MRI examinations. NeuroImage, 249, 118871. https://doi.org/10.1016/j.neuroimage.2022.118871