Granularity of Regional Homogeneity Changes in MDD (UKBB)

Jan 1, 2025 · 3 min read
projects research project

Project Overview

Major depressive disorder (MDD) is a prevalent psychiatric disorder with complex functional brain alterations. Standard brain atlases often dilute subtle disorder-specific effects in functional MRI analyses. This project introduces a data-driven, MDD-specific parcellation framework using Kullback-Leibler (KL) distance optimization to define cortical regions that maximize signal-to-noise for Regional Homogeneity (ReHo) measures in UK Biobank resting-state fMRI.

Original Article:
Exploring the Granularity of Illness-Related Changes in Regional Homogeneity in Major Depressive Disorder using UKBB Data — Pacific Symposium on Biocomputing (PSB) 2025 / World Scientific
DOI: 10.1142/9789819807024_0046

🔑 Key Highlights

  • Disorder-specific parcellation: KL distance is used to derive MDD-optimized cortical regions that increase contrast between cases and controls.
  • Enhanced RVI (Regional Vulnerability Index): Using MDD-specific regions produces larger effect sizes than standard anatomical atlases (e.g., Desikan).
  • Large-scale validation: Dataset includes 2,289 MDD cases and 6,104 healthy controls from UK Biobank.
  • Methodological innovation: Combines functional ReHo mapping with KL distance optimization for precise individual-level biomarkers.
  • Translational potential: Provides templates for functional biomarker extraction, stratification, and longitudinal monitoring.

🧠 Scientific Background

Functional brain alterations in MDD, particularly measured by ReHo (local BOLD signal synchrony), are heterogeneous and subtle. Standard atlas-based approaches average over large regions, reducing sensitivity. The adaptive KL parcellation framework:

  • Preserves illness-relevant signal variation
  • Maximizes statistical contrast for individual-level predictions
  • Can be generalized to other psychiatric or neurodegenerative disorders

🔬 Methodological Pipeline

Stepwise workflow:

  1. Preprocessing: Resting-state fMRI from UK Biobank processed (motion correction, normalization, filtering).
  2. ReHo computation: Local synchrony measured voxel-wise.
  3. KL parcellation: Derive data-driven cortical regions maximizing illness effect and SNR.
  4. RVI computation: Regional Vulnerability Index calculated with adaptive parcels vs standard atlases.
  5. Statistical validation: Large cohort evaluation demonstrates enhanced sensitivity of MDD-specific parcellations.

MDD ReHo KL Parcellation Pipeline

Conceptual workflow from raw fMRI to optimized MDD-specific ReHo biomarkers.

🧬 Multi-Modal & Translational Integration

While primarily focused on functional MRI, this framework can be extended to genetics, transcriptomics, and multi-omics data:

  • Integrate RNA-seq or GWAS data with ReHo-derived regions
  • Link functional vulnerability to molecular signatures
  • Enable mechanistic insights for precision psychiatry
  • Supports cross-disorder comparisons for schizophrenia, bipolar disorder, etc.

This multi-modal perspective enhances collaboration opportunities in computational psychiatry, functional genomics, and translational neuroscience.

🧠 Impact & Applications

  • Enables individualized ReHo biomarker extraction
  • Increases statistical power for translational studies
  • Provides template for adaptive parcellation in future studies
  • Facilitates clinical deployment for monitoring treatment response

👥 Authorship & Credits

This project is based on the PSB 2025 / World Scientific publication:

  • Yewen Huang, PhD – Conceptualization, algorithm design
  • Syed Ibrar Hussain, PhD – Computational modeling, statistical analysis
  • Paul M. Thompson, PhD – Neuroimaging genetics expertise
  • Peter Kochunov, PhD – Translational psychiatry & imaging
  • Additional Co-authors: Demetrio Labate, Robert Azencott, Bhim Adhikari

📄 Publication Reference

DOI: 10.1142/9789819807024_0046

🤝 Collaboration & Future Directions

  • Extend adaptive parcellation framework to multi-omics datasets (e.g., transcriptomics, proteomics)
  • Comparative atlas studies in schizophrenia, bipolar disorder, or anxiety
  • Individual-level RVI biomarkers for clinical research
  • Opportunities for collaborations in computational psychiatry, precision imaging, and multi-modal biomarker discovery
Major Depressive Disorder MDD Regional Homogeneity KL Distance UK Biobank Functional Neuroimaging Big Data Adaptive Parcellation Precision Psychiatry
Syed Ibrar Hussain, PhD
Authors
Syed Ibrar Hussain, PhD (he/him)
Senior AI Research Scientist
I am a postdoctoral researcher specializing in multimodal and generative deep learning for biomedical imaging and genetics. Experienced in developing CNNs, GANs, and diffusion models for medical imaging, integrating heterogeneous datasets to learn representations predictive of phenotypic and genotypic factors. Skilled in distributed HPC, Python-based deep learning frameworks (PyTorch/TensorFlow), and reproducible AI pipelines for research in high-dimensional biomedical data.