The 7T CMR workflow, end to end

How a 7T cardiac MRI moves from raw k-space to a radiologist-ready read: acquisition, AI motion-correction, QA gating, and the on-call paging path that only alerts when quality fails or a flagged finding needs review.

Acquisition

A 7T cardiac scan starts with a free-breathing, navigator-gated cine sequence tuned to the higher B0's susceptibility profile. We hold the TR short, push the bandwidth, and accept a longer scan in exchange for the SNR that 7T buys you. Every study is uploaded as a DICOM bundle — original cines, the recon series, and the technologist notes from the console.

Intake

The Intake agent reads the upload, normalizes the DICOM headers, and slots the study into the right cohort by protocol, field strength, and acquisition date. Studies that don't match a known protocol are parked in a manual-review lane with the discrepancy flagged — nothing is silently dropped.

Motion-corrected QA

The QA agent runs a learned motion-correct pass and grades each cine against a motion-blur + signal-homogeneity rubric. A clean cine moves straight to the read queue; a borderline cine flags with the suspected root cause (respiratory drift, arrhythmia gap, gating failure) so the radiologist sees the failure mode, not just a score.

Triage and paging

The Triage agent watches for two things only: QC failures a reviewer needs to weigh in on, and flagged findings that match the program's paging criteria. Everything else sits in the read queue. The on-call radiologist gets a single page with the cine slice, the QA verdict, and a deep link back to the full study — not fifty alerts a shift.

Read and close

The radiologist signs the report inside the dashboard, the report is pushed back to the hospital PACS, and the audit log is sealed. End-to-end latency is the metric the agents are tuned against — and the only metric we publish in the queue dashboard.