Chapter 26: Image Display and Enhancement

Here begins the image-processing part. Every part so far dealt with reconstruction: recovering an image from measurements. Once reconstruction is done, there is an image, or a volume. This part deals with what comes next: viewing, cleaning up, and measuring the image you obtained, the post-processing that leads to diagnosis or quantification. Whatever the scanner or modality, this downstream processing is largely shared. This first chapter starts with the processing for viewing.

CT numbers and windowing

CT pixel values are Hounsfield units (HU), a physical scale that fixes water at 0 and air at −1000. Bone exceeds +1000, fat is around −100, and soft tissues differ from one another by only tens of HU. The full range spans more than 2000, but a display can only distinguish so many gray levels at once, and the human eye sees only so many shades. So we cut out just the range of the tissue we want and map it from black to white. This is windowing.

Choosing a window width $\text{WW}$ and window level $\text{WL}$, the display value is

$$\text{display}(v) = \operatorname{clamp}\!\left(\frac{v - (\text{WL} - \text{WW}/2)}{\text{WW}},\ 0,\ 1\right)$$

Everything outside $[\text{WL}-\text{WW}/2,\ \text{WL}+\text{WW}/2]$ saturates to black (0) or white (1). The narrower the window, the steeper the transfer and the stronger the contrast within that narrow range. From the very same raw data, a soft-tissue window (WL 40 / WW 400), a bone window (WL 500 / WW 2000), and a lung window (WL −600 / WW 1500) reveal entirely different information. Note that windowing does not alter the underlying data; it changes only the display.

Simulation: the window changes which tissue you see

An HU phantom of representative tissues is shown under WL/WW control. Switch presets (soft, bone, lung, brain) or move the sliders. The middle histogram is overlaid with the transfer curve, showing which HU maps to which gray. Tissues outside the window collapse to black or white and can no longer be told apart. On the right is an enhanced image with histogram equalization applied within the current window.

Enhancement by histogram

The histogram of an image is the distribution of how many pixels have each brightness. A low-contrast image has its histogram bunched into a narrow range. Histogram equalization maps pixel values through the cumulative distribution function (CDF), spreading the used gray levels across the full range and raising contrast. Gamma correction $\text{out}=\text{in}^{\gamma}$ lifts the dark tones for $\gamma<1$ and the bright tones for $\gamma>1$, changing how tones are allotted nonlinearly. These are enhancements for viewing, and their purpose differs from measurement, where you want to preserve the original quantitative values (HU). The deep-learning image improvement of Chapter 13 is, broadly, in this lineage of enhancement, but it differs in that it learns the noise statistics and removes them.

Viewing a volume: MPR and MIP

CT acquires not a single slice but a volume of many thin stacked slices. There are two common ways to view it. Multiplanar reformation (MPR) re-slices the stacked volume along any plane (axial, coronal, sagittal, or oblique). Maximum intensity projection (MIP) takes the maximum along the line of sight and collapses it into one image, giving a through-depth overview of high-signal structures such as contrast-filled vessels. Both are display techniques made possible only once the reconstructed volume exists.

Key points

Because CT numbers (HU) span a wide range, windowing (WL/WW) assigns just the desired range from black to white for display. The window chosen changes which tissue you see, and this is a display operation that does not alter the data. Histogram equalization and gamma correction are enhancements that change how contrast is allotted, and MPR and MIP are reformatted displays for viewing volume data. That covers the processing for viewing an image. From the next chapter we move on to denoising to clean up noise, segmentation to extract structures, registration to overlay images, and finally quantification for measurement.

References

• Kalender WA. Computed Tomography: Fundamentals, System Technology, Image Quality, Applications, 3rd ed. Publicis (2011) — HU and windowing.
• Gonzalez RC, Woods RE. Digital Image Processing, 4th ed. Pearson (2018) — the standard text on histogram equalization, gamma, and enhancement.
• Fishman EK et al. Volume Rendering versus Maximum Intensity Projection in CT Angiography. RadioGraphics 26, 905–922 (2006).