Cine MR Images SEG Support (2D + t)

[1brahimmohamed]

I'm working with cardiac MRI cine sequences, which consist of multiple temporal frames at the same spatial location, each representing a different phase of the cardiac cycle. This is in contrast to conventional multi-frame CT or MR volumes where each frame corresponds to a different spatial slice.

When attempting to generate a segmentation for these cine sequences, I encounter the following error:

ValueError: Input image/frame positions are not unique according to the Dimension Index Pointers. The generated segmentation would be ambiguous. Ensure that source images/frames have distinct locations.

This error arises because all temporal frames share the same spatial position, and the library currently appears to use spatial position as the only differentiator among frames. However, cine MR data relies on temporal dimension information. such as the trigger time to differentiate between frames.

My question is:

• Is spatial position currently the only supported dimension for distinguishing frames in the segmentation process?
• If so, Is there any existing work around for using temporal dimension tags to differentiate frames when generating a segmentation?

[CPBridge]

Hi @1brahimmohamed thanks for asking this question. I'm afraid the simple answers are:

Is spatial position currently the only supported dimension for distinguishing frames in the segmentation process?

More or less.

If so, Is there any existing work around for using temporal dimension tags to differentiate frames when generating a segmentation?

No

In more detail. Currently the only supported options are:

• Spatial position using the Image Patient Position (for radiology images), plus the Segment Number for multiple segments
• Tiled spatial position using the {X,Y,Z} Offset in Slide Coordinate System (for pathology images), plus the Segment Number for multiple segments
• A single frame with no position information (something like an x-ray with no spatial information), plus the Segment Number for multiple segments

Others have requested other indexing schemes in the past with the addition of time being the most common (see #174 #200). It is very much on the roadmap. However rather than ad-hoc addition of new index dimensions, I want to refactor this in such a way that arbitrary (valid) dimensions can be specified by the user. Recent releases have been laying the groundwork for this, but it will take some further work. Unfortunately the number of possible interactions between the spatial and additional dimensions (and optionally segment number) makes this rather involved.

I will try and put this as a higher priority since you have indicated that it would be useful to you. If you are able to share any source images that could be used for testing, that would also be very helpful