Latent-Space Causal Discovery from Indirect Neuroimaging Observations

2026-01-30

AI Summary

Neuroimaging observations are confounded by hemodynamic responses and volume conduction, rendering it difficult to directly infer causal relationships among underlying neural variables. This work proposes a novel approach grounded in the physical invertibility of imaging modalities and the assumption of nonstationary latent dynamics. By applying a physics-aware inverse transformation to recover latent neural signals and integrating a delay-aware Mamba encoder to model time-lagged causal structures, the method explicitly accounts for temporal delays inherent in neuroimaging data. Crucially, it leverages mechanism shifts as sources of informational variation for causal discovery, establishes theoretical bounds on invertibility-induced error propagation, and enables zero-shot cross-task transfer validation. In The Virtual Brain (TVB) simulations, the approach achieves 2–3× higher F1 scores than baseline methods; applied to Human Connectome Project (HCP) motor-task fMRI data, it yields sparse directed graphs that prominently highlight canonical visuomotor pathways.

Abstract

Neuroimaging does not observe causal variables directly: hemodynamics and volume conduction distort signals so that statistical dependence need not reflect latent neural influence. Before estimating graphs, one must specify under what assumptions delayed directed structure can be studied from such indirect observations. We formalize a conditional setting - recoverable inversion under modality physics together with nonstationary latent dynamics - and derive an inversion-error propagation bound under explicit assumptions. Building on this framing, we propose INCAMA (INdirect CAusal MAmba): physics-aware inversion coupled with a delay-aware Mamba encoder that uses mechanism shifts as informative variation for directed graph scoring. We use controlled simulations for quantitative validation and HCP motor-task fMRI as a zero-shot external transfer check based on anatomical and task-network consistency. Across TVB simulations, INCAMA improves directed-structure recovery by 2-3x in F1 over observation-space and two-stage baselines, and on HCP motor-task fMRI it produces sparse directed estimates concentrated in canonical visuo-motor pathways.

Problem

• causal discovery
• neuroimaging
• latent variables
• indirect observations
• directed structure

Innovation

• causal discovery
• neuroimaging inversion
• nonstationary dynamics
• Mamba encoder
• mechanism shifts