What is a reconstruction kernel in CT?

A kernel (or filter) is the mathematical filter applied to the raw projection data during reconstruction. Soft kernels smooth the image and reduce noise, which suits soft tissue. Sharp kernels preserve edges and fine detail at the cost of more noise, which suits lung and bone.

What do window width and window level mean?

They decide how Hounsfield units are shown as grey. The level is the HU value at the middle of the grey scale; the width is the range of HU spread from black to white. A narrow window gives high contrast over a small range, such as the brain. A wide window shows a large range with less contrast, such as the lung or bone.

Why reconstruct thin slices if radiologists read thicker ones?

Thin slices keep the full spatial resolution in the z-axis, which is needed for good multiplanar reformats and 3D. Thicker slices average several thin ones, which reduces noise and makes the stack faster to scroll. Many protocols produce both from one acquisition.

Reconstruction, kernels and windows

One CT acquisition can become a soft-tissue series, a lung series and a bone series. The kernel, slice thickness and window decide which, and all three are chosen after the photons are already collected.

By the ScanFlo Editorial Team · Updated 6 October 2026

In this section · CT fundamentals

A CT acquisition does not produce an image. It produces projections: attenuation measured from hundreds of angles around the patient. The image you read is one of many that could be built from those measurements. Protocol basics lists the parameters; the aim here is to explain why the same raw data can give such different pictures.

Hounsfield units: the number under every pixel

Every reconstructed voxel holds an attenuation value on a fixed scale. Water is defined as 0 HU and air as −1000. Fat sits around −100, soft tissue a few tens above zero, and bone runs into the hundreds and beyond. Because the scale is fixed, the numbers mean something: a lesion measuring −90 HU contains fat, and a fluid collection that measures 40 HU is not simple fluid.

The kernel: sharpness against noise

Reconstruction applies a filter, called the kernel, to the projections before back-projecting them into the image. A soft kernel lets through mostly low spatial frequencies. Edges are gentle and noise is low, which is what soft tissue needs: a small difference in density against a quiet background. A sharp kernel lets high frequencies through. Fine structures such as bronchial walls, trabeculae and the ossicles become crisp, but the noise is amplified along with them.

Neither is better. They answer different questions from the same photons. That is why a chest CT usually leaves the console as a soft-tissue series and a lung series, and a trauma spine as soft tissue and bone. On the consoles ScanFlo follows, kernel names read as a letter and a number: B for body and H for head, with higher numbers sharper. A B30 is a soft body kernel; a B70 is a sharp one.

Iterative reconstruction, now standard on most scanners, models noise and the scanner geometry and refines the image over several passes. It lowers noise at a given dose, which is what makes lower-dose protocols usable. At high strength it can give soft tissue a smooth, slightly plastic texture that some readers dislike.

ScanFlo CT console Recon tab for a routine brain job: slice 4.0 mm, kernel H31s, window Soft Tissue, FoV 236 mm, recon begin −295 mm and end −445 mm, increment 4 mm, 39 images. — A reconstruction job in the ScanFlo CT console. Slice thickness, kernel and window are set per job, so one acquisition can be reconstructed as many times as the protocol needs.

Slice thickness and increment

Thin slices keep z-axis detail, show small nodules without partial-volume blurring and feed good reformats. They are also noisier, because each slice holds fewer photons. Thick slices average that noise away and are faster to review, but a 3 mm nodule in a 5 mm slice is diluted by its neighbours.

The usual answer is to reconstruct thin for the archive and for reformats, and thick (or thin-overlapping, then reformatted) for primary reading. The increment, the spacing between reconstructed slices, can be smaller than the thickness. Overlapping slices improve reformats and lesion detection at no dose cost.

The window: a second choice, made at the viewer

The kernel is baked into the series. The window is not. A CT image holds far more grey levels than a monitor or an eye can separate, so the viewer maps a chosen band of HU onto black-to-white. The level sets the centre of that band; the width sets how wide it is. Everything below the band is black, everything above is white.

Window	Width / Level (HU)	For
Brain	~80 / 40	Grey-white differentiation, early infarct
Soft tissue	~350–400 / 40–50	Abdomen, mediastinum
Lung	~1500 / −600	Parenchyma, airways
Bone	~1800–2000 / 400–500	Cortex, trabeculae, fractures

Kernel and window work together. A lung window on a soft-kernel series is usable but blurred; a soft-tissue window on a sharp-kernel series looks grainy. Match the pair to the question.

Try it in ScanFlo: Scan the chest case and reconstruct the same acquisition twice from the Recon tab: a soft kernel such as B30 at 5 mm and a sharp kernel such as B60 at 1 mm. View each with the lung and mediastinal presets. Then measure a region of the aorta on both series and compare the noise. See ScanFlo CT →

← Protocol basics Contrast timing and bolus tracking →