Matrix size, image resolution and count density

One of the things I do with the radiology residents is do some practical labs with them, where they get to see some of the things they've been learning about in the classes we give them before they take the first part of their board certification exams.

A few months ago, it was a demonstration on the gamma cameras in nuclear medicine, where I attempted to show them how changing the image matrix size affects the image.

Nuclear medicine images are typically acquired using matrix sizes ranging from 642 all the way up to 10242. Most are acquired using a 1282 image matrix size.

Generally, as you increase the matrix size, the resolution of the image improves. Most people might think that because of this, you should always use the highest matrix size possible, because that will give you the best image. But you'd be wrong.

Nuclear medicine images are acquired either for a fixed time, or a fixed number of counts. Let's suppose an image is acquired containing a total of 106 (1 million) counts. If the image matrix size is 642, that 106 counts is spread out over 4096 pixels (which gets you about 244 counts/pixel).

Below is an image of a bar quadrant phantom acquired in a 642 matrix for 106 counts.
64x64.png
It's a fairly smooth looking image, not a lot of noise, but none of the bars are resolved.

If that image matrix size is 1282 instead, now that million counts is spread out over 4 times as many pixels (16384 pixels), and now you only have 61 counts/pixel. Below is the same bar quadrant phantom, same 106 counts but acquired in a 1282 matrix size
128x128.png
Now one set of bars is visible but the noise has increased a little bit.

You can resolve smaller structures with the larger matrix size, but now the count density of your image has decreased by a factor of 4 (and the noise in your image has also increased).

Let's go to 2562
256x256.png
The next set of bars is visible but notice how the noise has increased significantly.

Let's see what 10242 looks like. You might expect that we would be able to see even more bars at this matrix size.
1024x1024.png
Nope, no more bars visible, and the noise has really gone up. At 10242, the count density has decreased to less than 1 count/pixel.

Of course there are ways around this. At higher matrix sizes, you can acquire more counts in your image but this requires either imaging with more activity in the patient or imaging for a longer time, which may or may not be feasible.

Photographers, take note. The same thing happens with digital cameras. Naturally, the amount of light digital cameras capture is orders of magnitude higher than in nuclear medicine, but problem is the same. More megapixels (MP) is not always better.