Haochang Shou

Imputation-Based Harmonization Mitigates Site Effects Without Data Leakage in Machine Learning Studies

The paper introduces MIRTH, an imputation-based harmonization method that mitigates site effects in neuroimaging data without data leakage, achieving high predictive accuracy in machine learning studies.

ComBatFamQC: Streamlining Interactive Batch-Effect Diagnostics and Harmonization for Neuroimaging Data in R

ComBatFamQC provides an R package for batch-effect diagnosis and harmonization in neuroimaging data, integrating visualization, statistical tests, and harmonization methods.

Batch Effect Correction for Neuroimaging Data with Heterogeneous Spatial Correlations

The paper develops and evaluates new batch effect correction methods for neuroimaging data that account for spatial correlations, showing improved performance over existing methods.

Mapping the spatiotemporal continuum of structural connectivity development across the human connectome in youth

The paper identifies a spatiotemporal continuum of structural connectivity development in youth, showing a shift from sensorimotor to association connectivity around age 15 and its links to cognition and psychopathology.

INTACT: A method for integration of longitudinal physical activity data from multiple sources

INTACT harmonizes time-varying physical activity data from multiple sources by removing device-specific effects while preserving biological variation, improving cross-study comparisons.

An open, fully-processed data resource for studying mood and sleep variability in the developing brain

The paper introduces an open data resource combining neuroimaging and behavioral measures to study mood and sleep in adolescents and young adults.

Neuroimaging-AI endophenotypes reveal underlying mechanisms and genetic factors contributing to progression and development of four brain disorders

The paper identifies nine neuroimaging endophenotypes linked to genetic loci and shows they improve prediction of systemic diseases and mortality.

DeepComBat: A Statistically Motivated, Hyperparameter-Robust, Deep Learning Approach to Harmonization of Neuroimaging Data

DeepCombat removes batch effects from neuroimaging data using a deep learning approach, outperforming existing methods in preserving biological variation.

pNet: A toolbox for personalized functional networks modeling

pNet is an open-source toolbox for creating personalized functional networks from fMRI data, offering multiple interfaces and methods to enhance reliability and reproducibility.

The Genetic Architecture of Multimodal Human Brain Age

This study identifies sixteen genome-wide significant loci associated with brain age gaps derived from MRI modalities and highlights potential causal relationships between brain aging and diseases.

Residual Partial Least Squares Learning: Brain Cortical Thickness Simultaneously Predicts Eight Non-pairwise-correlated Behavioural and Disease Outcomes in Alzheimer's Disease

This paper introduces re-PLS Learning to predict multiple AD outcomes from brain cortical thickness, identifying joint predictive regions across cohorts.

Similarity-Based Multimodal Regression

This paper introduces a distance-based regression method for multimodal data that can handle different data types and dimensionalities, showing its effectiveness through simulations and real-world studies.

Establishing the Validity of Compressed Sensing Diffusion Spectrum Imaging

This study evaluates six compressed sensing diffusion spectrum imaging (CS-DSI) schemes, finding they provide accurate and reliable white matter structure measures with up to 80% reduced scan time compared to full DSI.

Deconfounded Dimension Reduction via Partial Embeddings

The paper introduces the PARE framework to remove confounders from dimension reduction methods, showing it can eliminate batch effects in single-cell data and separate clinical from technical variability in neuroimaging.

Suboptimal phenotypic reliability impedes reproducible human neuroscience

Simulations and neuroimaging analyses argue that unreliable clinical and cognitive phenotypes, not only noisy fMRI, cap biomarker discovery; aggregating across raters or repeated measures beats enlarging samples.

Novel genomic loci and pathways influence patterns of structural covariance in the human brain

The paper identifies 617 novel genomic loci and pathways associated with structural covariance patterns in the human brain using a stochastic multivariate factorization method.

Harmonizing Functional Connectivity Reduces Scanner Effects in Community Detection

This study identifies scanner effects in brain community detection and proposes new harmonization methods to reduce these effects in functional connectivity analysis.

Multi-scale semi-supervised clustering of brain images: deriving disease subtypes

This paper introduces MAGIC, a multi-scale semi-supervised clustering method for brain imaging data to identify disease subtypes, and evaluates its performance on simulated and clinical datasets.

Privacy-preserving harmonization via distributed ComBat

This paper introduces distributed ComBat to harmonize neuroimaging data across institutions while preserving privacy, achieving results equivalent to centralized methods.

Removal of Scanner Effects in Covariance Improves Multivariate Pattern Analysis in Neuroimaging Data

This paper shows that removing scanner effects in mean and variance is insufficient for multivariate pattern analysis and proposes a method to harmonize covariance across scanners.

Three-level Sleep Stage Classification Based on Wrist-worn Accelerometry Data Alone

This paper develops an LSTM model to classify sleep stages using wrist-worn accelerometry data alone, achieving 65-81% accuracy with small training samples.

Statistical Pitfalls in Brain Age Analyses

The paper identifies statistical issues in brain age gap analysis, showing that modifying the brain age gap can inflate model accuracy statistics and suggests developing new methods to quantify deviation from normal developmental trajectories.

Harmonization of large multi-site imaging datasets: Application to 10,232 MRIs for the analysis of imaging patterns of structural brain change throughout the lifespan

The paper harmonizes 10,232 structural brain MRIs from 18 studies to create a normative reference of brain development and aging across 3 to 96 years.

The correspondence problem: which brain maps are significantly similar?

This paper introduces a spatial permutation framework to test the significance of overlap between brain maps and demonstrates its application to task-based activity, resting-state networks, and anatomical landmarks.