Comparison of individualized behavioral predictions across anatomical, diffusion and functional connectivity MRI

NeuroImage, Journal Year: 2022, Volume and Issue: 263, P. 119636 - 119636

Published: Sept. 16, 2022

A fundamental goal across the neurosciences is characterization of relationships linking brain anatomy, functioning, and behavior. Although various MRI modalities have been developed to probe these relationships, direct comparisons their ability predict behavior lacking. Here, we compared anatomical T1, diffusion functional (fMRI) at an individual level. Cortical thickness, area volume were extracted from T1 images. Diffusion Tensor Imaging (DTI) approximate Neurite Orientation Dispersion Density (NODDI) models fitted The resulting metrics projected Tract-Based Spatial Statistics (TBSS) skeleton. We also ran probabilistic tractography for images, which stream count, average length, each DTI NODDI metric tracts connecting pair regions. Functional connectivity (FC) was both task resting-state fMRI. Individualized prediction a wide range behavioral measures performed using kernel ridge regression, linear regression elastic net regression. Consistency results investigated with Human Connectome Project (HCP) Adolescent Brain Cognitive Development (ABCD) datasets. In datasets, FC-based gave best performance, regardless model or measure. This especially true cognitive component. Furthermore, all able cognition better than other components. Combining improved cognition, but not Finally, behaviors, combining resting FC yielded performance similar modalities. Overall, our study suggests that in case healthy children young adults, behaviorally-relevant information features might reflect subset variance captured by FC.

Language: Английский

---

Macroscopic resting-state brain dynamics are best described by linear models

Nature Biomedical Engineering, Journal Year: 2023, Volume and Issue: 8(1), P. 68 - 84

Published: Dec. 11, 2023

It is typically assumed that large networks of neurons exhibit a repertoire nonlinear behaviours. Here we challenge this assumption by leveraging mathematical models derived from measurements local field potentials via intracranial electroencephalography and whole-brain blood-oxygen-level-dependent brain activity functional magnetic resonance imaging. We used state-of-the-art linear families to describe spontaneous resting-state 700 participants in the Human Connectome Project 122 Restoring Active Memory project. found autoregressive provide best fit across both data types three performance metrics: predictive power, computational complexity extent residual dynamics unexplained model. To explain observation, show microscopic can be counteracted or masked four factors associated with macroscopic dynamics: averaging over space time, which are inherent aggregated activity, observation noise limited samples, stem technological limitations. therefore argue easier-to-interpret faithfully during conditions.

Language: Английский

---

Meta-matching as a simple framework to translate phenotypic predictive models from big to small data

Nature Neuroscience, Journal Year: 2022, Volume and Issue: 25(6), P. 795 - 804

Published: May 16, 2022

Language: Английский

---

Brain-age prediction: A systematic comparison of machine learning workflows

NeuroImage, Journal Year: 2023, Volume and Issue: 270, P. 119947 - 119947

Published: Feb. 16, 2023

The difference between age predicted using anatomical brain scans and chronological age, i.e., the brain-age delta, provides a proxy for atypical aging. Various data representations machine learning (ML) algorithms have been used estimation. However, how these choices compare on performance criteria important real-world applications, such as; (1) within-dataset accuracy, (2) cross-dataset generalization, (3) test-retest reliability, (4) longitudinal consistency, remains uncharacterized. We evaluated 128 workflows consisting of 16 feature derived from gray matter (GM) images eight ML with diverse inductive biases. Using four large neuroimaging databases covering adult lifespan (total N = 2953, 18-88 years), we followed systematic model selection procedure by sequentially applying stringent criteria. showed mean absolute error (MAE) 4.73-8.38 years, which 32 broadly sampled MAE 5.23-8.98 years. reliability consistency top 10 were comparable. choice representation algorithm both affected performance. Specifically, voxel-wise spaces (smoothed resampled), without principal components analysis, non-linear kernel-based performed well. Strikingly, correlation delta behavioral measures disagreed predictions. Application best-performing workflow ADNI sample significantly higher in Alzheimer's mild cognitive impairment patients compared to healthy controls. presence bias, estimates varied depending bias correction. Taken together, shows promise, but further evaluation improvements are needed its application.

Language: Английский

---

Graph Autoencoders for Embedding Learning in Brain Networks and Major Depressive Disorder Identification

IEEE Journal of Biomedical and Health Informatics, Journal Year: 2024, Volume and Issue: 28(3), P. 1644 - 1655

Published: Jan. 9, 2024

Brain functional connectivity (FC) networks inferred from magnetic resonance imaging (fMRI) have shown altered or aberrant brain connectome in various neuropsychiatric disorders. Recent application of deep neural to connectome-based classification mostly relies on traditional convolutional (CNNs) using input FCs a regular Euclidean grid learn spatial maps neglecting the topological information networks, leading potentially sub-optimal performance disorder identification. We propose novel graph learning framework that leverages non-Euclidean inherent structure for classifying major depressive (MDD). introduce autoencoder (GAE) architecture, built upon (GCNs), embed and node content large fMRI into low-dimensional representations. For constructing we employ Ledoit-Wolf (LDW) shrinkage method efficiently estimate high-dimensional FC metrics data. explore both supervised unsupervised techniques embedding learning. The resulting embeddings serve as feature inputs fully-connected network (FCNN) distinguish MDD healthy controls (HCs). Evaluating our model resting-state dataset, observe GAE-FCNN outperforms several state-of-the-art methods classification, achieving highest accuracy when LDW-FC edges features. also reveal significant group differences between HCs. Our demonstrates feasibility providing valuable discriminative diagnosing

Language: Английский

---

MRI economics: Balancing sample size and scan duration in brain wide association studies

bioRxiv (Cold Spring Harbor Laboratory), Journal Year: 2024, Volume and Issue: unknown

Published: Feb. 18, 2024

Abstract A pervasive dilemma in neuroimaging is whether to prioritize sample size or scan time given fixed resources. Here, we systematically investigate this trade-off the context of brain-wide association studies (BWAS) using functional magnetic resonance imaging (fMRI). We find that total duration (sample × per participant) robustly explains individual-level phenotypic prediction accuracy via a logarithmic model, suggesting and are broadly interchangeable up 20-30 min data. However, returns diminish relative size, which explain with principled theoretical derivations. When accounting for overhead costs associated each participant (e.g., recruitment, non-imaging measures), many small-scale some large-scale BWAS might benefit from longer than typically assumed. These results generalize across domains, scanners, acquisition protocols, racial groups, mental disorders, age as well resting-state task-state connectivity. Overall, our study emphasizes importance time, ignored standard power calculations. Standard calculations maximize at expense can result sub-optimal accuracies inefficient use Our empirically informed reference available future design: WEB_APPLICATION_LINK

Language: Английский

---

Predicting personality from network-based resting-state functional connectivity

Brain Structure and Function, Journal Year: 2018, Volume and Issue: 223(6), P. 2699 - 2719

Published: March 23, 2018

Language: Английский

---

Understanding Graph Isomorphism Network for rs-fMRI Functional Connectivity Analysis

Frontiers in Neuroscience, Journal Year: 2020, Volume and Issue: 14

Published: June 30, 2020

Graph neural networks (GNN) rely on graph operations that include network training for various related tasks. Recently, several attempts have been made to apply the GNNs functional magnetic resonance image (fMRI) data. Despite recent progresses, a common limitation is its difficulty explain classification results in neuroscientifically explainable way. Here, we develop framework analyzing fMRI data using Isomorphism Network (GIN), which was recently proposed as powerful GNN classification. One of important contributions this paper observation GIN dual representation convolutional (CNN) space where shift operation defined adjacency matrix. This understanding enables us exploit CNN-based saliency map techniques GNN, tailor with one-hot encoding, visualize regions brain. We validate our large-scale resting-state (rs-fMRI) classifying sex subject based structure The experiment consistent expectation such obtained show high correspondence previous neuroimaging evidences differences.

Language: Английский

---

Machine Learning With Neuroimaging: Evaluating Its Applications in Psychiatry

Biological Psychiatry Cognitive Neuroscience and Neuroimaging, Journal Year: 2019, Volume and Issue: 5(8), P. 791 - 798

Published: Nov. 27, 2019

Language: Английский

---

Multimodal brain-age prediction and cardiovascular risk: The Whitehall II MRI sub-study

NeuroImage, Journal Year: 2020, Volume and Issue: 222, P. 117292 - 117292

Published: Aug. 21, 2020

Brain age is becoming a widely applied imaging-based biomarker of neural aging and potential proxy for brain integrity health. We estimated multimodal modality-specific in the Whitehall II (WHII) MRI cohort using machine learning imaging-derived measures gray matter (GM) morphology, white microstructure (WM), resting state functional connectivity (FC). The results showed that prediction accuracy improved when multiple imaging modalities were included model (R2 = 0.30, 95% CI [0.24, 0.36]). GM WM models similar performance 0.22 [0.16, 0.27] R2 0.24 [0.18, 0.30], respectively), while FC lowest 0.002 [-0.005, 0.008]), indicating features less related to chronological compared structural measures. Follow-up analyses predictions similarly low matched sub-sample from UK Biobank, although consistently lower than predictions, with increasing sample size range. Cardiovascular risk factors, including high blood pressure, alcohol intake, stroke score, each associated WHII cohort. Blood pressure stronger association matter, no differences associations intake these observed. In conclusion, machine-learning based can reduce dimensionality neuroimaging data provide meaningful biomarkers individual aging. However, depends on study-specific characteristics range, which may cause discrepancies findings across studies.

Language: Английский

---