Interpretable, Multimodal, and White Matter-Inclusive Frameworks for Brain Network Biomarker Discovery in Aging and Neuropsychiatric Disease Open Access

Abstract

Understanding the brain’s complex functional architecture is critical for advancing diagnosis and treatment of neuropsychiatric disorders and age-related decline. However, the field faces two persistent challenges: the difficulty of interpreting machine learning models applied to brain connectivity, and the underexplored role of white matter (WM) in functional neuroimaging studies. This dissertation addresses both gaps through a sequence of studies that develop interpretable, multimodal frameworks to uncover novel brain network biomarkers across disorders and aging.  

First, we introduce an explainable machine learning pipeline leveraging SHapley Additive exPlanations (SHAP) to identify key functional network connectivity (FNC) features that distinguish schizophrenia patients from healthy controls, and young from older adults. Using Random Forest, XGBoost, and CATBoost models, our approach achieved 81.04% accuracy in schizophrenia classification and 71.38% accuracy in age group differentiation, while revealing the cognitive control, subcortical, and somatomotor networks as critical drivers. These findings highlight the need for multimodal approaches capable of capturing both functional and structural brain alterations. 

Motivated by this, our second study proposes a novel deep learning framework integrating resting-state fMRI spatial maps—via our voxelwise intensity projection (iVIP) method—with structural MRI to improve Alzheimer’s disease (AD) diagnosis. This multimodal 3D CNN not only surpasses traditional unimodal models (94.12% AD detection accuracy, 97.79 AUC) but also enables clinically relevant three-way differentiation among AD, mild cognitive impairment, and healthy aging groups. Saliency analyses reveal disease-relevant regions, underscoring the potential of fused modalities for early diagnosis. 

Yet, these approaches predominantly focus on gray matter (GM), overlooking WM’s emerging role in functional connectivity. To address this, our final study introduces the first large-scale WM intrinsic connectivity network (ICN) template, derived from over 100,000 fMRI scans. Application of this template to schizophrenia and task fMRI datasets reveals modular WM networks, robust task-related WM activation, and altered WM-GM connectivity, offering a new avenue for understanding brain disorders beyond GM-centric models. Together, these works provide interpretable, multimodal, and WM-inclusive frameworks that advance biomarker discovery and set the stage for more comprehensive models of brain dysfunction across aging and disease.

Table of Contents

Table of Contents

·       Chapter 1: Introduction and Background

o   1 Neuroimaging using MRI

o   2 Independent component analysis in functional neuroimaging analysis

o   3 Static functional connectivity in resting state fMRI

o   4 Functional brain network alterations across aging, schizophrenia and Alzheimer’s disease

·       Chapter 2: Visualizing Functional Network Connectivity Differences Using an Explainable Machine-learning Method

o   1 Introduction

o   2 Methods

o   3 Results

o   4 Discussion

o   5 References

·       Chapter 3: Voxel-wise Fusion of Resting fMRI Networks and Gray Matter Volume for Alzheimer's Disease Classification using Deep Multimodal Learning

o   1 Introduction

o   2 Methods

o   3 Results

o   4 Discussion

o   5 References

·       Chapter 4: Evidence for white matter intrinsic connectivity networks at rest and during a task: a large-scale study and templates

o   1 Introduction

o   2 Methods

o   3 Results

o   4 Discussion

o   5 References

·       Chapter 5: Conclusions and future directions

o   1 Conclusions

o   2 Discussions and future directions

·       Appendix A: Supplementary Tables

·       Appendix B: Supplementary Figures

·       Appendix C: Code and data availability statements

·       Funding acknowledgements

About this Dissertation

Primary PDF

Thumbnail	Title	Date Uploaded	Actions
	Interpretable, Multimodal, and White Matter-Inclusive Frameworks for Brain Network Biomarker Discovery in Aging and Neuropsychiatric Disease ()	2025-11-03 22:32:56 -0500	Press to Select an action Download

Supplemental Files