Goal: build a simulation engine that generates realistic MR imaging data
from a comprehensive biophysical representation of tissue (architecture +
parameters). This enables cheap synthetic training data for ve2e
(inverse) models, exploration of the space of possible MR images,
biophysical inductive biases, and MR sequence optimization.
Status at a glance (as of the last filed report, January 2026)
- Core simulator: sequence objects (pulseq-compatible), adaptive time-stepping, and an ensemble Kalman inversion (EKI) procedure for biophysical parameter estimation/calibration are implemented; EKI has been validated on toy domains, not yet on real tissue/phantom data.
- Multi-coil MRI reconstruction and imaging-noise modeling were theoretically systematized; a novel magnetization propagator for diffusion MRI was derived from the Feynman path-integral formalism.
- No report was filed for February 2026 (explicit “no report filed” stub);
no diffsim report exists for March 2026 either (no file at all — an
undocumented gap, unlike February’s explicit stub) — treat diffsim status
as frozen at January 2026 until a newer report or
inbox/session shows up.
Priorities (per the last Q1 plan)
- Ship a first usable simulator version + basic verification/validation/UQ (VVUQ, per the ASME standard).
- Acquire and scan an ex-vivo phantom; finalize EKI parameter estimation on real data.
- Run joint histology–MRI + simulator sensitivity-analysis experiments
(coordination point with the
histoworkstream on domain-representation labeling requirements).
Source material
- sources/monthly-reports/ —
filed as
<MM>-2026_diffsim.md. Only January 2026 has real content; Feb 2026 is an explicit “no report filed” stub; no file at all exists for Mar 2026 onward (undocumented gap, not a stub).
No dedicated projects/ or sops/ subdirectories exist yet for this
workstream — add them here as diffsim work gets synthesized from inbox/.