Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Recovering Whole-Brain Causal Connectivity under Indirect Observation with Applications to Human EEG and fMRI

Sangyoon Bae, Miruna Oprescu, David Keetae Park, Shinjae Yoo, Jiook Cha

TL;DR

This work tackles the challenge of recovering whole-brain directed connectivity from indirect EEG and fMRI observations by introducing INCAMA, a framework that jointly performs physics-aware inversion and latent-space causal discovery. By modeling modality-specific observation processes (e.g., HRF deconvolution for fMRI and leadfield mixing for EEG) and leveraging nonstationary regime shifts as informative interventions, INCAMA achieves identifiability and robust causal graph recovery in the latent neural space. The approach demonstrates strong performance on large-scale simulations and zero-shot generalization to real HCP motor-task fMRI, recovering canonical visuo-motor pathways with high consistency to known neuroanatomy, while remaining computationally efficient. These results establish a scalable, theoretically grounded blueprint for causal inference from indirect, distorted neuroimaging data, with potential to guide brain-network analyses in research and clinical contexts.

Abstract

Inferring directed connectivity from neuroimaging is an ill-posed inverse problem: recorded signals are distorted by hemodynamic filtering and volume conduction, which can mask true neural interactions. Many existing methods conflate these observation artifacts with genuine neural influence, risking spurious causal graphs driven by the measurement process. We introduce INCAMA (INdirect CAusal MAmba), a latent-space causal discovery framework that explicitly accounts for measurement physics to separate neural dynamics from indirect observations. INCAMA integrates a physics-aware inversion module with a nonstationarity-driven, delay-sensitive causal discovery model based on selective state-space sequences. Leveraging nonstationary mechanism shifts as soft interventions, we establish identifiability of delayed causal structure from indirect measurements and a stability bound that quantifies how inversion error affects graph recovery. We validate INCAMA on large-scale biophysical simulations across EEG and fMRI, where it significantly outperforms standard pipelines. We further demonstrate zero-shot generalization to real-world fMRI from the Human Connectome Project: without domain-specific fine-tuning, INCAMA recovers canonical visuo-motor pathways (e.g., $V1 \to V2$ and $M1 \leftrightarrow S1$) consistent with established neuroanatomy, supporting its use for whole-brain causal inference.

Recovering Whole-Brain Causal Connectivity under Indirect Observation with Applications to Human EEG and fMRI

TL;DR

This work tackles the challenge of recovering whole-brain directed connectivity from indirect EEG and fMRI observations by introducing INCAMA, a framework that jointly performs physics-aware inversion and latent-space causal discovery. By modeling modality-specific observation processes (e.g., HRF deconvolution for fMRI and leadfield mixing for EEG) and leveraging nonstationary regime shifts as informative interventions, INCAMA achieves identifiability and robust causal graph recovery in the latent neural space. The approach demonstrates strong performance on large-scale simulations and zero-shot generalization to real HCP motor-task fMRI, recovering canonical visuo-motor pathways with high consistency to known neuroanatomy, while remaining computationally efficient. These results establish a scalable, theoretically grounded blueprint for causal inference from indirect, distorted neuroimaging data, with potential to guide brain-network analyses in research and clinical contexts.

Abstract

Inferring directed connectivity from neuroimaging is an ill-posed inverse problem: recorded signals are distorted by hemodynamic filtering and volume conduction, which can mask true neural interactions. Many existing methods conflate these observation artifacts with genuine neural influence, risking spurious causal graphs driven by the measurement process. We introduce INCAMA (INdirect CAusal MAmba), a latent-space causal discovery framework that explicitly accounts for measurement physics to separate neural dynamics from indirect observations. INCAMA integrates a physics-aware inversion module with a nonstationarity-driven, delay-sensitive causal discovery model based on selective state-space sequences. Leveraging nonstationary mechanism shifts as soft interventions, we establish identifiability of delayed causal structure from indirect measurements and a stability bound that quantifies how inversion error affects graph recovery. We validate INCAMA on large-scale biophysical simulations across EEG and fMRI, where it significantly outperforms standard pipelines. We further demonstrate zero-shot generalization to real-world fMRI from the Human Connectome Project: without domain-specific fine-tuning, INCAMA recovers canonical visuo-motor pathways (e.g., and ) consistent with established neuroanatomy, supporting its use for whole-brain causal inference.
Paper Structure (94 sections, 5 theorems, 71 equations, 2 figures, 30 tables, 2 algorithms)

This paper contains 94 sections, 5 theorems, 71 equations, 2 figures, 30 tables, 2 algorithms.

Table of Contents

  1. Introduction
  2. Background and Related Work
  3. Problem Setting: Causal Discovery from Indirect Human Brain Measurements
  4. Latent Delayed Dynamical SCM
  5. Indirect Observation Model
  6. Estimand and Learning Objective
  7. Identifiability under Indirect Observation and Nonstationarity
  8. Assumptions
  9. Main Results (proofs deferred)
  10. INCAMA: Indirect Causal Mamba
  11. Physics-aware Inversion
  12. Latent-Space Causal Discovery
  13. Training Objective and End-to-End Optimization
  14. Experiments
  15. Experimental Setup
  16. ...and 79 more sections

Key Result

Theorem 4.6

Consider the model class where (i) $\mathbf{z}_{1:T}$ is generated by a delayed dynamical SCM satisfying assum:no_instantaneous--assum:latent_identifiable_class, and (ii) observations follow $\mathbf{x}_t=\mathcal{H}_\psi(\mathbf{z}_{t-L:t})+\bm{\epsilon}_t$ for which assum:invertibility holds. Then

Figures (2)

  • Figure 1: Overview of INCAMA.(a) End-to-end framework.INCAMA decomposes causal discovery from indirect neuroimaging measurements into physics-aware inversion and latent-space causal inference, mapping observed fMRI and EEG signals to latent neural dynamics before estimating directed connectivity. (b) Latent-space causal discovery. Directed, delay-aware causal inference is performed on latent neural trajectories using a scalable state-space sequence model with Mamba-based temporal encoding, yielding sparse and interpretable whole-brain connectivity. (c) physics-aware inversion. Modality-specific inversion modules, including HRF deconvolution for fMRI and source localization for EEG, recover latent neural activity from indirect and spatially mixed observations.
  • Figure 2: Recovered average visuo-motor causal pathways in the (a) left and (b) right hemispheres. Nodes correspond to standard neuroanatomical regions of interest (ROIs) defined by the Desikan--Killiany atlas. Arrows indicate the directional information flow inferred by INCAMA from real fMRI data of the HCP S1200 Motor Task, computed by averaging recovered causal graphs across 1,078 subjects. The resulting group-level structure reveals a consistent hierarchical progression from visual processing (V1 $\rightarrow$ V2), to motor planning (SMA $\rightarrow$ PM), and execution--feedback loops (M1 $\leftrightarrow$ S1).

Theorems & Definitions (5)

  • Theorem 4.6: Identifiability from indirect observations via inversion
  • Proposition 4.7: Inversion error propagation
  • Corollary 4.8: Consistency of the end-to-end estimator
  • Lemma 2.1: Vanishing reconstruction error implies vanishing test-function gap
  • Lemma 2.2: Top-$k$ support stability under a margin