iterative reconstruction
Family of CT reconstruction methods that iteratively refine the image to reduce noise at the cost of more computation per slice. Modeled on physics priors (system geometry, statistical noise, beam hardening) rather than the analytic single-pass approach of filtered back projection (FBP). Superseded FBP as the clinical default in the mid-2010s for dose reduction; deep-learning reconstruction (TrueFidelity / DLIR, AiCE) is the next-generation evolution.
Vendor-named variants:
- GE — ASiR (1st gen) → ASiR-V (model-based) → TrueFidelity / DLIR (deep learning).
- Siemens — IRIS → SAFIRE → ADMIRE.
- Philips — iDose → IMR → AI Suite.
- Canon — AIDR 3D → AiCE (deep learning).
Why it matters to buyers: Industry-standard since the mid-2010s. License tier is the largest software refurb price-determining variable on CT. A scanner without current-tier recon licensing performs at the older platform's dose / image-quality curve regardless of chassis age.
Why it matters to engineers: Compute-heavy — reconstruction time longer than FBP, particularly on early-generation iterative implementations. Modern systems use GPU acceleration to keep recon close to real-time. License-tier upgrades sometimes require recon-hardware upgrades, not just software keys.