ASiR-V and Deep Learning Reconstruction on GE CT: Settings That Matter

Iterative reconstruction and deep learning reconstruction are now standard on many CT scanners, but they are also among the most misunderstood settings at the console. We regularly see protocols where ASiR-V is switched on at a high percentage "for cleaner images" while the tube current hasn't changed at all, or where a deep learning series is missing because it was set on the wrong recon. This guide walks through how these reconstructions work on GE Revolution EVO systems, how they interact with Noise Index and automatic mA, and the common mistakes to avoid. The concepts apply to other manufacturers' iterative and deep learning reconstructions too, even though the names, strength scales and menus differ.

What these reconstructions actually do

Filtered back projection (FBP) is the classic reconstruction. It is fast and predictable, but image noise rises quickly as dose drops.

Iterative reconstruction (GE's current version is ASiR-V) models noise and the system and works to remove it. On the Revolution EVO, ASiR-V is selected as a percentage from 10% to 100% in 10% steps. The final image is a blend of the original image and a fully noise-reduced image, so a higher percentage means more noise reduction. GE's documentation states that spatial resolution (MTF) is maintained as the level increases. Images are annotated with the level, for example "AR40", in the upper left corner.

Deep learning image reconstruction (DLIR), marketed by GE as TrueFidelity, uses a trained neural network to separate noise from signal. It is offered at three strengths: Low, Medium and High. GE positions it as removing noise while keeping a noise texture closer to the FBP look technologists and radiologists are used to. Images are annotated "DL" plus the strength.

The practical difference that radiologists notice is texture. A 2021 RadioGraphics review describes iterative images at strong settings as looking blotchy or pixelated, and notes that low-contrast detection is not always preserved when dose is cut aggressively. A 2022 study comparing DLIR with ASiR-V reported lower noise and a more natural texture with DLIR, particularly in soft tissue.

The biggest misconception: turning on ASiR-V does not lower the dose

This is the one we teach at the scanner most often. Selecting an ASiR-V level does not change mA or Noise Index on its own. If you switch ASiR-V on and change nothing else, you get the same dose with less noise.

To use iterative reconstruction for dose reduction, one of these has to happen:

• Raise the Noise Index (when using AutomA/SmartmA), or lower the mA (manual technique) on the mA control screen, or
• Use Dose Reduction Guidance (DRG). On the Revolution EVO you enter a percentage dose reduction, and the system calculates the new mA and selects an ASiR-V level for you. The manual states a maximum DRG value of 50%; beyond that, DRG is set to 0% and adjustments are made manually.

A few related points from the manufacturer documentation:

• GE's dose management guide notes that Noise Index usually needs to change by at least about 15% to see a visible difference, and that dose varies with the inverse square of noise. Small NI tweaks may change dose without visibly changing the images.
• Applying DRG to a protocol that has already been dose-reduced can produce sub-optimal or non-diagnostic images. Know your starting point.
• If you change the system-recommended ASiR-V level or turn ASiR-V off, the system warns you. Treat that as a prompt to re-check mA and Noise Index before you scan.
• Cutting dose without regard to body habitus risks photon starvation and streak artifacts, which no reconstruction fully fixes.

Any change to technique factors or Noise Index belongs with your radiologist and medical physicist. GE's own guidance is that the site team should evaluate image quality at several levels and techniques before changing clinical protocols, then make changes in steps and review the results.

Setting it up at the console (Revolution EVO)

ASiR-V

1. On the Recon tab, open Recon Options and set ASiR-V to On, then choose the percentage.
2. ASiR-V can be applied prospectively on Recon 1 or on additional prospective recons, or retrospectively through Retro Recon.
3. For retrospective ASiR-V, create a new series so the iterative images are easy to identify and compare against the original.

DLIR

1. DLIR is set in Recon Options (DLIR Setup On), but only on additional recons (R2 onward) or Retro Recon, not on Recon 1. This trips people up: if you build DLIR into Recon 1, it won't be there.
2. GE recommends Medium as the starting strength, after review by your physicist and radiologist.
3. DLIR is available with the Standard kernel for axial and helical scanning (Standard and Detail for cardiac), over a defined range of slice thicknesses. In helical mode, the image interval must match the manual's rules for the chosen thickness, so check your recon spacing if DLIR greys out.
4. DLIR is an option on the Revolution EVO. It is not available on the Optima CT660.

Watch the order of your recons

GE's dose guide points out that AutomA uses Recon 1's slice thickness to set tube current. Keep Recon 1 as the thickest slice you intend to read, and add thinner, ASiR-V or DLIR series as extra recons.

Where these reconstructions don't apply

Check compatibility before you build a protocol around them. According to the Revolution EVO documentation:

• ASiR-V does not work with real-time modes such as SmartPrep (bolus tracking monitor images), SmartView, SmartStep and biopsy modes. It is also not compatible with the cardiac noise-reduction filters, and Temporal Enhance is unavailable at high ASiR-V levels.
• DLIR is not compatible with several modes, including SmartPrep, SmartView and SmartStep, Volume Helical Shuttle, narrow-collimation axial scans, axial overlapped recon and very short scan ranges. It is not validated for calcium scoring or quantitative perfusion, so keep those on their validated reconstruction.

Image quality cautions to know

• Small, low-contrast structures: GE notes that Medium and High DLIR can reduce the contrast of small structures. If that's a concern for an exam type, a Low-strength reconstruction is the documented fallback. Talk to your radiologist about which exams warrant it.
• Patterned noise: With very low-noise data, DLIR can show a patterned noise appearance. Lower strength, or a slightly wider soft-tissue window (liver, brain), reduces it.
• Metal: DLIR reduces partial-volume effects, which can change how metal looks. Compare against a non-DLIR series when hardware is in the field.
• Lung: The same 2022 comparison found DLIR's advantage over ASiR-V was large in soft tissue but small in lung. Don't assume the same gain everywhere.

Quick checklist before you save a protocol

• Recon 1 is the thickest slice you'll read (AutomA reference).
• If ASiR-V is meant to reduce dose, Noise Index or mA has actually been changed (or DRG used), with physicist and radiologist sign-off.
• DLIR is on an additional recon or retro recon, not Recon 1.
• Kernel, slice thickness and interval are DLIR-compatible.
• Calcium scoring and perfusion use their validated reconstruction.
• Patient is centered in the gantry, since centering errors throw off automatic mA regardless of reconstruction.
• You've confirmed the steps in the operator manual for your system and software version.

Key takeaways

• ASiR-V is a blend from 10% to 100%; more percentage means more noise reduction, not less dose.
• Dose only drops when Noise Index or mA changes, manually or through Dose Reduction Guidance.
• DLIR (Low, Medium, High) goes on additional or retro recons, with Medium as GE's suggested start.
• Neither works with real-time modes like SmartPrep, and DLIR isn't validated for calcium scoring or perfusion.
• Every protocol change goes through your radiologist and physicist, made in steps and reviewed.

Steps above apply to the GE Revolution EVO as described in its operator manual addendum. Menus and options vary by system and software version, so confirm in the manual for your own scanner. Have a reconstruction question we didn't cover? Reply and tell us what you're seeing, or request applications training for your team.

FAQ

Does turning on ASiR-V lower the CT dose?

No. Selecting an ASiR-V level on its own does not change mA or Noise Index, so the dose stays the same and the images are less noisy. Dose only drops when Noise Index or mA is changed, manually or with Dose Reduction Guidance.

What ASiR-V percentage should I use?

GE's documentation does not recommend one level. It advises that your physicist and radiologist review image quality at several levels and techniques first, then make changes in steps.

Why can't I select deep learning reconstruction on Recon 1?

On the Revolution EVO, DLIR is only available on additional recons (R2 onward) and Retro Recon. Add it as an extra series and keep Recon 1 as your thickest reference slice.

Can I use DLIR for calcium scoring?

No. GE states DLIR is not validated for calcium scoring or quantitative perfusion, so keep those exams on their validated reconstruction.

Sources

• Mohammadinejad P, et al. CT Noise-Reduction Methods for Lower-Dose Scanning: Strengths and Weaknesses of Iterative Reconstruction Algorithms and New Techniques. RadioGraphics, 2021.
• The Value of Deep Learning Image Reconstruction in Improving the Quality of Low-Dose Chest CT Images. Diagnostics, 2022.
• Manufacturer operator documentation: GE Revolution EVO / Optima CT660 Operator Manual Addendum (5974408-1EN, Rev. 2).
• Manufacturer documentation: GE Blueprint, A Guide to CT Radiation Dose Management.