CausalMamba: Scalable Conditional State Space Models for Neural Causal Inference

2025-10-20

AI Summary

This study addresses two key bottlenecks in fMRI-based neural causal inference: (1) the ill-posed inverse problem arising from hemodynamic distortion of BOLD signals, and (2) the computational intractability of methods such as Dynamic Causal Modeling (DCM). We propose a scalable two-stage framework: first, jointly deconvolving BOLD signals and estimating latent neural activity via a conditional state-space model; second, performing causal graph inference using a novel conditional Mamba architecture—enabling efficient long-range dependency modeling and capturing dynamic network reconfiguration. On synthetic data, our method achieves a 37% improvement in causal identification accuracy over DCM. Applied to real task-based fMRI, it reconstructs known neuroanatomical pathways with 88% fidelity—substantially outperforming conventional approaches, which fail in >99% of subjects. Moreover, our framework reveals, for the first time, the flexible switching of causal hubs during working memory—a fundamental mechanism of adaptive brain function.

Technology Category

• Machine Learning: Causal Learning
• Cognitive Modeling & Cognitive Systems: Neural Spike Coding
• Reasoning under Uncertainty: Causality

Application Category

• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Social Networks and Social Media: Computational social science
• Responsible Web: Machine-in-the-loop, human agency and autonomy

Abstract

We introduce CausalMamba, a scalable framework that addresses fundamental limitations in fMRI-based causal inference: the ill-posed nature of inferring neural causality from hemodynamically distorted BOLD signals and the computational intractability of existing methods like Dynamic Causal Modeling (DCM). Our approach decomposes this complex inverse problem into two tractable stages: BOLD deconvolution to recover latent neural activity, followed by causal graph inference using a novel Conditional Mamba architecture. On simulated data, CausalMamba achieves 37% higher accuracy than DCM. Critically, when applied to real task fMRI data, our method recovers well-established neural pathways with 88% fidelity, whereas conventional approaches fail to identify these canonical circuits in over 99% of subjects. Furthermore, our network analysis of working memory data reveals that the brain strategically shifts its primary causal hub-recruiting executive or salience networks depending on the stimulus-a sophisticated reconfiguration that remains undetected by traditional methods. This work provides neuroscientists with a practical tool for large-scale causal inference that captures both fundamental circuit motifs and flexible network dynamics underlying cognitive function.

Problem

• Inferring neural causality from hemodynamically distorted fMRI signals
• Addressing computational intractability of existing causal inference methods
• Capturing brain's dynamic causal network reconfigurations during cognition

Innovation

• Decomposes causal inference into two tractable stages
• Uses BOLD deconvolution to recover latent neural activity
• Employs Conditional Mamba architecture for causal graph inference