Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Behavior Score Prediction in Resting-State Functional MRI by Deep State Space Modeling

Javier Salazar Cavazos, Maximillian Egan, Krisanne Litinas, Benjamin Hampstead, Scott Peltier

TL;DR

This paper investigates predicting Alzheimer's-related behavior scores from resting-state fMRI by modeling temporal dynamics with a deep state-space framework. The authors introduce NeuroMamba, a bidirectional, differential, sparsity-enforcing architecture that processes multivariate BOLD timeseries to forecast MoCA, memory, and language scores, outperforming connectivity-based and other timeseries baselines. Across LOOCV experiments, NeuroMamba achieves the strongest correlations (MoCA R ≈ 0.36) and reveals biologically plausible predictive regions within the DMN and memory networks; ablations quantify the contribution of each design, with domain adaptation showing strong few-shot transfer to another dataset (ADNI). These results highlight the importance of temporal information in rs-fMRI for early cognitive impairment assessment and suggest practical avenues for cross-dataset deployment and targeted interventions. The work provides a foundation for integrating deep temporal models into neuroimaging-based risk monitoring and intervention strategies while offering interpretability through regionwise feature importance.

Abstract

Early clinical assessment of Alzheimer's disease relies on behavior scores that measure a subject's language, memory, and cognitive skills. On the medical imaging side, functional magnetic resonance imaging has provided invaluable insights into the neural pathways underlying Alzheimer's disease. While prior studies have used resting-state functional MRI by extracting functional connectivity matrices, these approaches neglect the temporal dynamics inherent in functional data. In this work, we present a deep state space modeling framework that directly leverages the blood-oxygenation-level-dependent time series to learn a sparse collection of brain regions to predict behavior scores. Our model extracts temporal features that encapsulate nuanced patterns of intrinsic brain activity, thereby enhancing predictive performance compared to traditional connectivity methods. We identify specific brain regions that are most predictive of cognitive impairment through experiments on data provided by the Michigan Alzheimer's Disease Research Center, providing new insights into the neural substrates of early Alzheimer's pathology. These findings have important implications for the possible development of risk monitoring and intervention strategies in Alzheimer's disease.

Behavior Score Prediction in Resting-State Functional MRI by Deep State Space Modeling

TL;DR

This paper investigates predicting Alzheimer's-related behavior scores from resting-state fMRI by modeling temporal dynamics with a deep state-space framework. The authors introduce NeuroMamba, a bidirectional, differential, sparsity-enforcing architecture that processes multivariate BOLD timeseries to forecast MoCA, memory, and language scores, outperforming connectivity-based and other timeseries baselines. Across LOOCV experiments, NeuroMamba achieves the strongest correlations (MoCA R ≈ 0.36) and reveals biologically plausible predictive regions within the DMN and memory networks; ablations quantify the contribution of each design, with domain adaptation showing strong few-shot transfer to another dataset (ADNI). These results highlight the importance of temporal information in rs-fMRI for early cognitive impairment assessment and suggest practical avenues for cross-dataset deployment and targeted interventions. The work provides a foundation for integrating deep temporal models into neuroimaging-based risk monitoring and intervention strategies while offering interpretability through regionwise feature importance.

Abstract

Early clinical assessment of Alzheimer's disease relies on behavior scores that measure a subject's language, memory, and cognitive skills. On the medical imaging side, functional magnetic resonance imaging has provided invaluable insights into the neural pathways underlying Alzheimer's disease. While prior studies have used resting-state functional MRI by extracting functional connectivity matrices, these approaches neglect the temporal dynamics inherent in functional data. In this work, we present a deep state space modeling framework that directly leverages the blood-oxygenation-level-dependent time series to learn a sparse collection of brain regions to predict behavior scores. Our model extracts temporal features that encapsulate nuanced patterns of intrinsic brain activity, thereby enhancing predictive performance compared to traditional connectivity methods. We identify specific brain regions that are most predictive of cognitive impairment through experiments on data provided by the Michigan Alzheimer's Disease Research Center, providing new insights into the neural substrates of early Alzheimer's pathology. These findings have important implications for the possible development of risk monitoring and intervention strategies in Alzheimer's disease.
Paper Structure (39 sections, 12 equations, 4 figures, 7 tables)

This paper contains 39 sections, 12 equations, 4 figures, 7 tables.

Table of Contents

  1. Introduction
  2. Background
  3. Related Works
  4. Contributions & Motivation
  5. Data Acquisition & Processing
  6. Dataset
  7. Behavior Score Metrics
  8. Preprocessing
  9. Formulation & Related Methods
  10. Problem Formulation
  11. Functional Connectivity
  12. Functional Connectivity Matrix (FCM)
  13. Kernel Ridge Regression (KRR)
  14. Model-Based Timeseries
  15. Individual Independent Component Analysis (I-ICA)
  16. ...and 24 more sections

Figures (4)

  • Figure 1: Violin plots illustrating the distribution of behavioral scores in z-score space across different disease categories.
  • Figure 2: Overview of the proposed NeuroMamba architecture for behavior score prediction using deep state space modeling. The Mamba++ layer extracts temporal features from each brain region relevant to prediction. Temporal averaging is subsequently applied to derive a single scalar summary statistic per region, which is then processed by a linear head.
  • Figure 3: Correlation scatter plots displaying the relationship between predicted NeuroMamba scores and true behavioral metrics (rows), across the Alzheimer’s disease spectrum (columns), presented in z-score normalized space.
  • Figure 4: Receiver operating characteristic (ROC) curve with area under curve (AUC) values for diagnosis between cognitively normal and non-normal subjects.