Window and Level on CT: What Radiologists Do With the Mouse

I'm sure you've sat next to a radiologist and watched them jerk the mouse around with small movements that alter the image on the screen. You might be wondering what they're doing and why. Those movements are windowing, the controls that decide which part of the density scale your screen actually shows.

These controls are especially difficult when you first start looking at CTs yourself, and it’s understandably easiest to default to the presets and hope. So let’s build the whole mechanism from the ground up, and then come back to the presets and decode them.

The Problem Windowing Solves#

A CT scanner measures how much each small block of tissue attenuates the X-ray beam. When we call something dense on CT, that’s the claim we’re making: this material knocks down the beam more effectively than its neighbors. The Hounsfield scale standardizes that measurement, anchored to water at 0 and air at -1000, and it runs past +2000 for cortical bone and higher for metal. Call it four thousand meaningfully different values inside a single head CT.

Your monitor can only display a few hundred shades of gray, and your eye can only reliably distinguish a few dozen. So how do we represent four thousand values? We can’t, at least not all at once. Something has to decide which densities get a gray of their own and which get lumped together, and that something is the window.

Before defining it, it helps to know where the interesting tissues actually sit on the scale:

Air

-1000 HU. Paranasal sinuses, mastoids, and any air that should not be there.

Fat

Around -100 HU. Orbital fat, scalp fat.

Water

0 HU by definition. CSF sits just above it.

Brain

White matter around 25 HU, gray matter around 35 to 40. A ten-unit stretch carries the entire gray-white interface.

Acute blood

50 to 100 HU. Clot is retracted, protein-packed plasma, so it out-attenuates the brain around it.

Bone

Several hundred to a few thousand HU depending on how cortical it is.

Read the brain entry again. The gray-white interface you assess on every head CT lives inside a ten-unit stretch of a four-thousand-unit scale. Most of the display problem in neuroimaging comes down to giving that stretch enough gray levels to be readable.

Window Width and Window Level, Defined#

Windowing is zoom, but for density instead of position. Window level is where you point: the HU value at the center of your display, rendered as middle gray. Window width is how far you zoom: the range of HU values spread across the full ramp from black to white. Anything below the bottom edge of the window renders pure black, anything above the top edge renders pure white, and no detail survives outside the window no matter how large the density difference.

A narrow window maps a small stretch of the scale across the full black-to-white ramp, so a difference of a few HU becomes a visible difference in gray. A wide window stretches that same ramp over thousands of units, so nothing clips at either end, but neighboring tissues land on shades too close to tell apart.

That is the whole trade, and you make it every time you touch the mouse.

Watch the Trade-Off Happen#

Put up a head CT in the standard brain window, W80 L40. The window runs from 0 to 80 HU: CSF near black, white matter dark gray, gray matter lighter, acute blood approaching white. The ten-unit gray-white difference covers about an eighth of the ramp, which is plenty to see. Meanwhile the calvarium, at many hundreds of units, sits off the top of the window and renders as a featureless white band.

Now watch what happens as the width opens up toward bone settings. The skull develops texture: cortex, diploë, sutures, fracture lines. At the same time the brain converges on a single flat gray, because its whole ten-unit interface now occupies a fraction of one displayed shade. Nothing about the data changed. The window is mapping a different part of the scale, and the brain fell outside the part that gets any contrast.

If you like seeing the arithmetic, the whole mechanism fits in a few lines:

def gray(hu: int, ww: int, wl: int) -> int:
    """Map a Hounsfield value to a display gray, 0-255."""
    lo = wl - ww / 2
    return round(255 * min(1.0, max(0.0, (hu - lo) / ww)))

# Acute subdural blood (65 HU) next to cortex (38 HU):
gray(65, 80, 40)     # brain window -> 207
gray(38, 80, 40)     # brain window -> 121   (86 grays apart: obvious)
gray(65, 2800, 600)  # bone window  -> 79
gray(38, 2800, 600)  # bone window  -> 76    (3 grays apart: invisible)

The same two voxels, 27 HU apart, are separated by 86 gray levels in one window and 3 in the other. The data is identical in both. Whether you see the blood depends entirely on which window you chose.

SYNAXIALSOFT TISSUEImage 24/48

SOFT TISSUEINVERTED

Which CT Window Preset for Which Question#

Each preset is a named position in the same trade-off: more contrast for one question, less for everything else.

The subdural preset W200 L80 is worth understanding rather than memorizing. It raises the level toward clot density and opens the width just enough that bone and blood stop sharing pure white, which is exactly the failure that hides the thin subdural at brain settings.

Viewers put these on hotkeys, usually the number row, and pair them with a free drag: hold W (or the middle button, viewers disagree) and drag one axis for width, the other for level. Learn the presets until you can hit them without looking. Then learn the free drag, because it lets you interrogate a specific pixel instead of hoping a preset covers it.

Reading a Head CT With This#

The practical sequence: brain window for the parenchyma, subdural window along the convexities and falx, bone window for the fractures, soft tissue for the scalp. The scalp hematoma tells you where the coup is, and the contrecoup you should look hardest for sits opposite it.

When a finding is equivocal at every preset, stop flipping windows and measure it. An ROI reads the data directly, with no gray ramp in the way.