why prefer isotropic over anisotropic for medical image datasets?

refer ResampleVolumesToBeIsotropic.cxx

I am new to medical image processing.

I can understand that visualization on a plane can be distorted for anisotropic data, but I cannot appreciate “I do not think 3D is informative” in the following from the referenced file.

Shouldn’t a signal processing algorithm take into account unequal pixel spacings in e.g., xy and z directions?

Also, is there a situation where I must have anisotropic data?

//The abundance of datasets acquired with anisotropic voxel sizes bespeaks a
// dearth of understanding of the third dimension and its importance for
// medical image analysis in clinical settings and radiology reading rooms.
// `Datasets acquired with large anisotropies bring with them the regressive
// message: {I do not think 3D is informative}.
// They stubbornly insist: {``all that you need to know, can be known
// by looking at individual slices, one by one’'}. However, the fallacy of
// this statement is made evident by simply viewing the slices when
// reconstructed in any of the orthogonal planes. The rectangular pixel shape
// is ugly and distorted, and cripples any signal processing algorithm not
// designed specifically for this type of image.

That statement say that whoever acquired an image with large anisotropy (e.g. 1x1x5, or 0.5x0.5x3) implies “I do not think 3D is informative”. While 3D is quite informative. Many tools and image processing algorithms expect isotropic data, so this example is provided.

1 Like