Segmentation: quantifying signal in each spot

The process called segmentation is when the software converts the images of each spot or feature and converts them to a series of numbers (intensities) that will be used to calculate the ratios. This sounds simple, but as with many aspects of DNA microarrays, it can get complex. MAGIC Tool helps you understand three different segmentation methods.

After you have gridded the slide (telling the software where the spots are located), you need to help the software understand what is signal and what is background. When MAGIC Tool draws a red circle, it is indicating all pixels insided the circle are considered signal and all pixels outside the red circle are considered background.

Fixed Circle: MACIC Tool puts the red circle in the middle of the yellow squares you drew for gridding. This is the fastest and a pretty reliable method. It is a good default if you have done a decent job of gridding. If your gridding was off a bit, the spot may not be centered in the grid box and therefore the signal will not be completely within the red circle (see figure above). You can adjust the radius of this circle as shown in the three figures below. Note that even if the red circle is bigger than the yellow box, only signal inside the yellow box and inside the red circle is used for measuring signal. Only pixels inside the yellow box but outside the red circle will be considered background, regardless of the brightness of the pixels.

Fixed Circle with Radius of 3 Pixels	Fixed Circle with Radius of 6 Pixels	Fixed Circle with Radius of 8 Pixels

Adaptive Circle: The size and the location of the circle changes, depending of the size on the feature on the microarray. See the instructors guide for more details on this algorithm. As you can see in the image below, the adapted circle does a better job of getting most of the signal inside the red circle and keeping most of the background outside the red circle. This is particularly important if you want to subtract background or average the pixels inside the red circle.

You can adjust the threshold of the adaptive circle, as well as the radius limits of the red circle. Depending on the particualr spot, a low threshold may do better at finding the spot (below left) than a high threshold setting (below right) even if the radii are set to exactly 6 pixels.

Adaptive Circle with Radius of 6 pixels, Threshold 25	Adaptive Circle with Radius of 6 pixels, Threshold 75

Seeded Region Growing: This method is designed to find the signal for each spot based on the distribution of the signal. This method for segmentation looks for the brightest pixel near the center of the grid square (the seed), and then connects all pixels (region growing) adjacent to this seed pixel and connects them into one shape.

The algorithm simultaneously connects pixels to background and foreground regions, continuing until all pixels are in one of the regions. A user-specified threshold determines which pixels can be used to “seed” the regions. This is the slowest method since each pixel is processed individually. The bigger the threshold, typically the bigger the spot will be defined. In the example below, you can see some of the problems with seeded region growing because the red signal (left) only includes two pixels while the exact same spot for the green signal (right) includes nearly every spot. This is a problem when the signals are low and can be filtered out later by excluding spots below a threshold of intensity which you can determine. Imagine the impact on this ratio using this method. What would happen if you averaged the pixels or subtracted background?

Seeded Region Growing Red Signal (faint signal)	Seeded Region Growing Green Signal (faint signal)

 

As with the adaptive circle, you can adjust the threshold and affect the number of pixels included as signal, or excluded as backgroud.

Seeded Region Growing (High Threshold)	Seeded Region Growing (Low Threshold)

In general, it is good to experiment with the different segmentation methods. You should also experiment with the consequences of whether you use total signal (all the pixel intensities added together) or averaged signal (total signal intensity divided by the number of signal pixels). Both of these methods can be performed with or without subtracting background signal.