CT Reconstruction Kernels: Choosing Soft Tissue, Bone and Lung

Pick the wrong kernel and a good scan can still look wrong: lungs that look smeared, a liver full of grain, a fracture line that fades into the background. Choosing CT reconstruction kernels well is one of the quickest image quality wins at the console. It adds no dose, and as long as the scan data is still on the system, a missed series can often be rebuilt after the patient has left the table. Here is how to match kernels to soft tissue, bone and lung, and build them into your protocols.

What a reconstruction kernel actually does

A kernel, also called a convolution algorithm or filter, is applied to the projection data before back projection. Radiopaedia's article on CT kernels explains that it reshapes the frequency content of the data to counter blurring, which is what turns simple back projection into filtered back projection.

The part that matters at the scanner is the trade-off. In the Image Wisely overview of reconstruction techniques, Mayo Clinic physicists Lifeng Yu and Shuai Leng explain that every kernel balances spatial resolution against noise: smoother kernels lower noise but blur fine detail, and sharper kernels show more detail but add noise.

A 2023 study in Diagnostics shows the same pattern in clinical images. The kernel sharpness study in photon-counting coronary CTA found that as kernel sharpness increased, image noise and vessel sharpness both rose while contrast-to-noise ratio steadily fell.

Matching the kernel to the task

Soft tissue: smooth to standard

Image Wisely notes that smooth kernels suit brain exams and liver lesion assessment, where lower noise helps you see subtle, low-contrast differences. For head, neck soft tissue, abdomen and pelvis, a smooth or standard kernel is the usual primary series.

Bone: sharp

Both Image Wisely and Radiopaedia describe bone kernels as sharper, higher-resolution reconstructions. Use them for fractures and cortical or trabecular detail, and expect grainy soft tissue on that series.

Lung: sharp, with one caveat

Lung parenchyma has high natural contrast against air, so a sharp lung kernel brings out fine detail and the extra noise matters less. The caveat is quantitative software. A 2023 Respiration study of emphysema scoring reconstructed 98 COPD scans with smooth, medium smooth and sharp kernels and found the emphysema score shifted significantly with the kernel: the sharp kernel read on average 6.2 percentage points higher than the smooth one, and one patient's score varied by up to 18% across kernels. If your site sends chest CT for emphysema quantification, use the kernel the analysis protocol specifies and keep it the same for every follow-up.

Reading kernel names on your scanner

Kernel names are not standardized across manufacturers. On Siemens systems the name is a letter code plus a number. In the Diagnostics study above, Br stood for body regular, Bv for body vascular and Qr for quantum regular, and the number (36, 40, 44, 48) was the sharpness level, with higher numbers sharper. GE systems label recon types by name instead; GE's Revolution EVO and Optima CT660 documentation refers to types such as Standard and Detail.

Because the naming and the noise behavior differ, do not assume a kernel on one system matches a similar-sounding kernel on another. If you are standardizing protocols across a mixed fleet, compare the images side by side and involve your physicist before you call two kernels equivalent.

Setting it up at the console

The steps below follow GE's Revolution EVO / Optima CT660 user manual addendum. Other systems have equivalent recon job screens, so confirm the details in the manual for your own system and software version.

1. Build the series into the protocol. The Recon tab supports a primary recon plus additional prospective recons (R2 to R10), so a chest protocol can produce its soft tissue, lung and bone series automatically instead of relying on someone to remember them.
2. Fix a missed series with Retro Recon. Open Retro Recon, choose the exam and series, and assign a new series number. GE's addendum advises keeping start and end locations, slice thickness, interval, display field of view and centering matched to the original series, then changing the recon type.
3. Check how iterative and deep learning recon interact with the kernel. ASiR-V is set in Recon Options in 10% steps, and the manual notes its dose reduction guidance is optimized for the Standard type. On the Revolution EVO, deep learning recon (DLIR) is limited to Standard for axial and helical scans, with Standard and Detail for cardiac, so a bone or lung series in that protocol needs a different recon setup. Our post on ASiR-V and deep learning reconstruction settings covers those options in detail.
4. Read the annotation. GE marks ASiR-V images with "AR" plus the level and DLIR images with "DL-" plus the strength, so you can confirm at a glance what a series was built with.

Image Wisely's advice is to choose the kernel and slice thickness together for each clinical application, using the lowest dose that still answers the question. Any protocol change should go through your radiologist and physicist.

Kernel checklist before you send the study

• Every series in the protocol has a named kernel and slice thickness
• Soft tissue, lung and bone series are all present when the order calls for them
• Each series is displayed in the matching window (soft tissue, lung or bone)
• Missing series are rebuilt before the scan data is removed from the system
• Quantitative chest follow-ups use the same kernel as the baseline

Key takeaways

• Every kernel trades spatial resolution against noise; sharper means more detail and more noise.
• Use smooth to standard kernels for soft tissue, sharp kernels for bone and lung.
• Kernel choice changes quantitative results, so keep it fixed for emphysema scoring and follow-ups.
• Kernel names differ by manufacturer; compare images before treating two as equivalent.
• Build every needed series into the protocol and use retro recon to fix misses while the data is still available.

FAQ

What is a reconstruction kernel in CT?

It is the filter applied to the projection data during filtered back projection. It sets the balance between sharpness and noise in the final images.

Does a sharper kernel increase radiation dose?

No. The kernel is applied after the data is acquired, so it changes image noise, not the dose delivered.

Can I change the kernel after the scan?

Usually yes. Most systems can reconstruct a new series from the scan data with a different kernel, as long as that data has not been removed from the system.

Which kernel is best for lungs?

A sharp lung kernel is the usual choice for viewing lung parenchyma. For emphysema quantification, use the kernel your analysis protocol specifies, because scores change with the kernel.

Are kernel names the same on every scanner?

No. Each manufacturer uses its own naming and the images behave differently, so check your system's manual and compare images when matching protocols across scanners.

Questions about kernels on your own scanners? Reply or reach out, or request applications training and we will work through your recon sets with your team at the console.

Sources

• Yu L, Leng S. Image Reconstruction Techniques. Image Wisely (ACR, RSNA, ASRT, AAPM), updated 2016.
• Murphy A. Kernel (image reconstruction for CT). Radiopaedia, revised 2023.
• Yang Y, et al. Optimization of Kernel Type and Sharpness Level Improves Objective and Subjective Image Quality for High-Pitch Photon Counting Coronary CT Angiography. Diagnostics, 2023.
• Bakker JT, et al. Effect of Chest Computed Tomography Kernel Use on Emphysema Score in Severe COPD Patients Evaluated for Lung Volume Reduction. Respiration, 2023.
• Manufacturer operator documentation: GE Revolution EVO / Optima CT660 User Manual Addendum (5974408-1EN, Rev 2).