Zooming and interpolating medical images

Hello,

This question is for someone with experience in displaying medical images (i.e., DICOMS).

Many DICOM viewers perform some interpolations on the images before displaying them. This prevents the displayed images from looking overly pixelated. Does anyone know what kind of interpolations are generally used for DICOM viewers (e.g., linear, cubic, spline, etc...)?

Is there MATLAB support for implementing any of these interpolation schemes when displaying images using imtool or imshow?

Thanks, Justin

Answer 1

[John Crane]

interp() (and variants) seems to work fine for what you are talking about. Many medical imaging applications like cubic spline, bi-cubic, but all the popular methods are used really. Generally, radiologists prefer the utmost Lossless methods that are possible. 1st of all, the possibility of any artifacts will (and should) automatically disqualify a method. In medical imaging we would rather see a low resolution illustration, including low SNR or even machine artifact included if that is the case, rather than the possibility of any artificial influence of an algorithm onto the data. You must be careful while we still have scanners at low resolutions (as compared to other technology). You can have an influence that shows in only 1 pixel or voxel on the acquired data, and, that thing will turn into a seemingly large "significant" area by over-interpolating.

imtool is not the most efficient way to observe images. I would suggest imshow() or similar methods in Matlab, with your own optimization.

[Justin Solomon]

Thanks for you insight. I'm actually quite amazed that the DICOM standard doesn't seem to say anything about interpolating. Since medical images are hardly ever displayed at their native size (i.e., 1 image pixel per displayed pixel), the interpolation scheme plays a large role in how the displayed image appears. I can't find any papers in which someone has compared different interpolation methods in terms of detectability of subtle lesions.