Segmentation Thresholds and Gating

The Segmentation Thresholds and Gating Window

The `Segmentation Thresholds and Gating' panel allows you to control Stradx's semi-automatic segmentation facilities. At the top of the panel are two buttons that allow you to select either standard ultrasound images, or thresholded images controlled by the sliders below. To see the standard ultrasound images, click on the `Grey' button (or the `Doppler' button, if you have recorded a colour Doppler data set). To see the thresholded data, click on the `Sliders' button.

When `Sliders' is selected, all Stradx's display windows show a modified image, where some pixels are highlighted in magenta. Which pixels come out as magenta depends on the settings of the six sliders lower down the `Segmentation Thresholds and Gating' panel. For gresycale (not Doppler) data, only the first pair of `Grey' sliders is effective. Suppose you want to highlight only those pixels with a greyscale intensity between 30 and 80: just set the `Low' slider to 30 and the `High' slider to 80. Immediately, you'll see all these pixels highlighted in magenta in all Stradx's display windows, with all other pixels unchanged. If the sliders are crossed, so that `Low' is higher than `High', then you'll highlight only those pixels with a greyscale value above `Low', as well as those with a greyscale value below `High'. When the sliders are crossed to the extremes, as in the example above, no pixels are highlighted.

This facility is useful if you are dealing with data that can be segmented by thresholding. Just set the `Low' and `High' sliders so that you see the structure you wish to segment highlighted in magenta, with as little background clutter as possible. Segmentation by thresholding is not feasible in many ultrasound applications, with the notable exception of colour Doppler data (see below). Note that selecting `Sliders' affects the appearance of all Stradx display windows, including the `Preview' window. This allows you to see the thresholded image before you record the data set and, perhaps, adjust the ultrasound machine's controls (the TGC curve, for instance) to improve the segmentation.

The Segmentation Thresholds and Gating Window

With colour Doppler data sets, the `Red' and `Blue' sliders control which coloured pixels to highlight alongside the selected greyscale pixels. In the example above, Stradx has been asked to highlight only the red pixels in the data set.

The Review Window

Above is a B-scan from a colour Doppler data set, displayed in the `Review' window. This is what Stradx displays by default. However, if `Sliders' is selected, with only the red pixels being highlighted (as above), then the `Review' window changes to display the image below.

The Review Window

Segmentation of the blood vessels can now be carried out relatively easily, without having to trace carefully around each magenta blob. Instead, Stradx allows you to just click on each blob with the left mouse button (a simple click without moving the mouse, otherwise Stradx thinks you are starting to manually trace a segmentation contour). Stradx uses a grass-fire algorithm to locate the boundary of the selected blob. The analogy is that a fire spreads out from the clicked-on pixel, igniting each magenta (flammable) pixel in the fire's path, until the edge of the flammable material (magenta pixels) is reached. Note that this facility is only available with thresholded images selected using the `Sliders' button.

The Review Window

Segmentation contours are automatically fitted to the edge of the burned-out region, as shown above. These are standard Stradx segmentation contours, so they can be edited using the middle mouse button or deleted using the right mouse button. They show up as coloured curves in the `Outline' window, affect the planimetry volume estimate and contribute to any surfaces displayed in the `Surface' window. Note that it is easy to click accidentally on a blob twice, producing two coincident segmentation contours, which might result in unexpected volume estimates or surface renderings. Stradx draws your attention to this potential problem by displaying coincident segmentation contours in a much thicker line width (in both the `Review' and `Outline' windows). Use the right mouse button to delete an unwanted contour.

To cope with small amounts of noise in the thresholded image, Stradx provides one control to assist the grass-fire algorithm. The `Fire jump' slider, located at the top of the `Segmentation Thresholds and Gating' panel, controls how far a fire will jump over unhighlighted (not flammable) pixels to reach magenta (flammable) pixels on the other side. If there is a lot of noise in the thresholded data, then it might be helpful to increase this control above the default value of 2. This will give you smoother segmentation contours with fewer small holes in the middle of large blobs. As well as filling in small holes, the `Fire jump' control also filters out small blobs, so you will not be able to segment tiny magenta regions unless you set `Fire jump' to a small value: think of `Fire jump' as a general resolution control.

The thresholding process, followed by the grass-fire procedure, provides a quick and robust way to segment colour Doppler data (and, in certain cases, greyscale data). The four blood vessels above were segmented in about one second using four mouse clicks. The image below shows the segmentation contours superimposed on the original, colour Doppler data (just click on the `Doppler' button to see this).

The Review Window

A full 3D segmentation requires segmentation of more than one B-scan. You can either scan through the data set manually, clicking on the magenta blobs in pertinent B-scans, or use one of Stradx's more automated segmentation facilities. For example, you can automatically pick out all the magenta blobs in every 20th frame using the `Go' button in the `Segmentation Thresholds and Gating' panel. If you have the `Outline' window open, and have clicked on `All' to see all the B-scans, you'll see the blobs appearing as Stradx detects them. This facility takes as its starting point the frame currently displayed in the `Review' window: it segments this frame, then every 20th frame earlier and later in the sequence. If you'd like to segment a different number of frames, for example every 30th frame, simply adjust the slider next to the `Go' button before pressing `Go'. Note that the `Fire jump' control is still effective, so you'll get smoother contours for higher settings of `Fire jump'.

The `Segmentation Thresholds and Gating' panel also allows you to segment and visualise parts of the body moving under the influence of the cardiac pulse.