[Paper Review] A Dual-Head Transformer-State-Space Architecture for Neurocircuit Mechanism Decomposition from fMRI

Cole Korponay|arXiv (Cornell University)|Jan 20, 2026

Functional Brain Connectivity Studies0 citations

TL;DR

Introduces Neurocircuit Mechanism Decomposition (NMD), a dual-head model combining a graph-constrained lag-aware transformer with a measurement-aware state-space model to decompose fMRI functional connectivity into drive, input responsivity, and modulator gating for targeted treatment insights.

ABSTRACT

Precision psychiatry aspires to elucidate brain-based biomarkers of psychopathology to bolster disease risk assessment and treatment development. To this end, functional magnetic resonance imaging (fMRI) has helped triangulate brain circuits whose functional features are correlated with or even predictive of forms of psychopathology. Yet, fMRI biomarkers to date remain largely descriptive identifiers of where, rather than how, neurobiology is aberrant, limiting their utility for guiding treatment. We present a method for decomposing fMRI-based functional connectivity (FC) into constituent biomechanisms - output drive, input responsivity, modulator gating - with clearer alignment to differentiable therapeutic interventions. Neurocircuit mechanism decomposition (NMD) integrates (i) a graph-constrained, lag-aware transformer to estimate directed, pathway-specific routing distributions and drive signals, with (ii) a measurement-aware state-space model (SSM) that models hemodynamic convolution and recovers intrinsic latent dynamics. This dual-head architecture yields interpretable circuit parameters that may provide a more direct bridge from fMRI to treatment strategy selection. We instantiate the model in an anatomically and electrophysiologically well-defined circuit: the cortico-basal ganglia-thalamo-cortical loop.

Motivation & Objective

• Motivate precision psychiatry by linking fMRI biomarkers to actionable neurobi circuit mechanisms.
• Decompose fMRI functional connectivity into output drive, input responsivity, and modulator gating.

Proposed method

• Develop a graph-constrained, lag-aware transformer to estimate directed, pathway-specific routing distributions and drive signals.
• Integrate with a measurement-aware state-space model to capture hemodynamic convolution and recover latent dynamics.
• Provide an interpretable set of circuit parameters bridging fMRI data and treatment strategy.

Experimental results

Research questions

• RQ1 How can fMRI-based functional connectivity be decomposed into identifiable neurocircuit mechanisms?
• RQ2 Can a dual-head architecture yield interpretable, mechanism-level parameters aligned with therapeutic interventions?
• RQ3 What is the role of a lag-aware transformer and a measurement-aware SSM in recovering intrinsic neural dynamics from fMRI?
• RQ4 How does the approach apply to the cortico-basal ganglia-thalamo-cortical loop?

Key findings

• Proposes a dual-head architecture that yields interpretable circuit parameters for neurocircuit mechanism decomposition.
• Integrates a graph-constrained lag-aware transformer with a measurement-aware state-space model to model hemodynamics and latent dynamics.
• Demonstrates instantiation in a well-defined cortico-basal ganglia-thalamo-cortical loop.
• Offers a framework to move beyond descriptive biomarkers toward mechanism-informed treatment guidance.