neurojax-phantom-synthetic

name: neurojax_phantom_synthetic description: Quantify Inverse Solver Leakage and Resolution using a Synthetic Phantom.

Synthetic Phantom Validation

This agent validates the resolution properties of the neurojax inverse solvers (Native ADMM/Scico) by computing the Resolution Matrix and quantifying leakage artifacts.

Objectives

1. Generate Synthetic Source Space: Create a random or grid-based source space and corresponding random Leadfield $L$.
2. Compute Resolution Matrix: Calculate $R = G \cdot L$ for the Native ADMM solver.
3. Quantify Leakage:
   • Peak Localization Error (PLE): Distance between the peak of the PSF (row of R) and the true source.
   • Spatial Dispersion (SD): Spread of the PSF around the peak.
   • Crosstalk: Magnitude of off-diagonal elements in the CTF (column of R).
4. Visualize: Plot PSF and CTF for representative sources.

Instructions

1. Create/Run examples/demo_phantom_synthetic.py.
2. The script must:
   • Simulate a setup with $N_{sensors} \approx 64$ and $N_{sources} \approx 256$.
   • Run inverse_scico.solve_inverse_admm to get the inverse operator $G$ (or approximate it via linearity if using L1).
   • Note: For non-linear solvers (L1), $R$ is strictly state-dependent. Use small perturbations or a linear approximation (Weighted MNE) for stable Resolution Matrix analysis, OR empirically compute PSF by inverting unit impulses $\delta_i$.
   • Save plots to results_phantom_synthetic/.
3. Report statistical metrics (Mean/Max PLE).