Stephen M. Smith

DHARM is an open-source tool that diagnoses, harmonizes, and evaluates batch effects in MRI data using standardized workflows and multivariate diagnostics.

This study investigates the drivers of dynamic functional connectivity in resting-state fMRI using HMMs and finds that mean activity and variance dynamics explain most of the apparent connectivity changes.

This paper identifies latent subgroups in functional brain organisation using large-scale resting-state fMRI data from UK Biobank participants, revealing associations with cognitive, lifestyle, and health phenotypes.

This paper introduces a framework to model brain connectivity by separating functional segregation and gradients, showing it better approximates the connectome than alternative approaches.

This paper addresses bias in cortical thickness estimation in UK Biobank by using harmonisation methods to correct for batch effects caused by missing T2-FLAIR data.

The paper compares individual variability in fMRI and MEG neural fingerprints, finding low correspondence at the subject level despite population-level similarity.

This paper introduces bi-cross-validation to evaluate and compare dynamic functional connectivity models in fMRI, showing it balances model fit and complexity effectively.

The paper defends lesion network mapping by showing that it provides meaningful differences through specificity testing, refuting criticisms that it reflects normative connectome properties.

The paper examines how age affects the relationship between brain structure and cognition, finding that models trained on younger subjects predict cognition in older subjects better than vice versa.

This paper develops a 24-channel 3T RF coil for fMRI in awake monkeys, improving ferumoxytol-weighted functional connectivity estimation and identifying human-like functional networks.

This study examines how brain volume differences in adults are stable from young adulthood or change with age, using MRI data from 8,311 individuals.

This paper introduces Independent Filter Analysis (IFA), a supervised method for group discrimination in fMRI that outperforms traditional ICA in detecting group differences.

This paper introduces an automated individual cortical parcellation method using a novel weakly supervised classifier, improving performance over previous models and identifying new atypical language-related brain areas.

The paper disambiguates ongoing brain aging from baseline effects using cross-sectional data and validates findings with longitudinal data from UK Biobank.

The paper describes the release of a large fetal fMRI dataset and presents methods for its preprocessing and analysis to study in utero brain connectivity.

This paper introduces mPFMs, a new method for multiscale functional brain connectivity analysis that improves prediction of UK Biobank traits compared to standard techniques.

This study identifies 66 genetic variants linked to neuroendocrine structure volumes and links some to testosterone levels using UK Biobank data.

This paper introduces a transfer learning method to accurately localise the Vim in clinical-quality MRI data by transferring anatomical information from high-quality datasets.

This paper introduces an automated tICA pipeline for fMRI data to remove artifacts and improve signal detection, achieving high accuracy in component classification.

This paper constructs internally-consistent, fully-unbiased multimodal MRI brain templates from UK Biobank data, showing improved registration accuracy compared to existing templates.

The paper improves prediction of cognitive traits from fMRI data by stacking multiple dynamic brain network models to reduce variability and enhance accuracy.

MMORF is a nonlinear image registration tool for MRI brain images that outperforms existing methods in accuracy and distortion control across multiple domains.

This paper evaluates individual differences in brain network organization and identifies spatial overlap as indicative of linear additive coupling.

The paper estimates brain age gap (BAG) using neural networks and identifies eight genomic regions associated with BAG, finding causal relationships with Alzheimer's and bipolar disorder but not Parkinson's, depression, or schizophrenia.

The paper uses neuroimaging data from the UK Biobank to analyze functional connectivity between interoceptive brain regions and links it to health-related domains using canonical correlation analysis.

The paper investigates structural and functional asymmetry in the neonatal cerebral cortex using MRI data from term and preterm neonates and adults.

This paper generalises the HCP multimodal cortical parcellation to UK Biobank data using a geometric deep learning framework, improving detection rates across cortical areas.

This paper introduces a semi-supervised, multimodal, and multi-task fusion method for discovering image-derived phenotypes from brain imaging data, showing improved prediction accuracy compared to conventional approaches.

The paper improves predictions of individual differences in task-evoked fMRI activity using a sparse ensemble learner and shows better performance than previous methods.

This paper improves the interpretability of brain-behavior CCA by using domain-driven dimension reduction, showing more stable and interpretable results compared to standard PCA.