Support parallel annotations of multiple sequences

[kbressem]

Is your feature request related to a problem? Please describe. We are working on a project where the interpretation of a pathological finding depends on the context of multiple MRI sequences. Specifically we need a T2 weighted fat saturated sequence and a T1 weighted sequence. Sometimes a third sequence might even be needed. Currently MONAILabel and Slicer do only support annotations of a single 3D volume.

Describe the solution you'd like Annotation of multiple sequences, ideally in different windows that are synced with each other.

Describe alternatives you've considered One can manually load images to ITK-SNAP. If multiple ITK-SNAP windows are opened, it will sync across the sequences. This is our current solution, but requires a lot of manual effort.

Additional context This is our current workflow with an example segmentation.

The images are synced across the two ITK-SNAP windows, the segmentations are done manually.

[diazandr3s]

Hi @kbressem,

I fully agree this feature is very much needed for several use cases: Mutlisequence and multilable tasks.

As far I understand, this is mostly a GUI feature - MONAI Label (server itself) has all APIs needed for this.

We should also add here that all sequences should be aligned or co-registered and of the same resolution?

Please let us know your thoughts.

[kbressem]

Hi @diazandr3s, yes it is mostly an GUI issue. I believe it would be beneficial if the image volumes can be aligned, but it should not be mandatory. I am also not sure if one should do just a single annotation that is mapped to all sequences, or of the possibility for multiple annotation should be allowed. I see usecases for both. Mapping a segmentation to multiple images saves time, but there might be alignment issues. Sometimes, especially in out current project, certain pathologies are only seen in a single MRI sequence, but to correctly interpret them, additional information from other sequences is needed.

[diazandr3s]

Thanks for clarifying, @kbressem. I think BRATS is one of the use cases - all modalities are aligned and some regions are better highlighted in certain sequences/modalities. I'd suggest starting with the assumption that the sequences are aligned. Some viewers have registration algos available

[StijnvWijn]

Hello @diazandr3s and @kbressem , I am currently encountering the same problem with the SPIDER dataset. It has multiple sequences per subject, but I can't seem to get MONAI Label to recognize them as belonging to the same subject. I have tried the following things:

1. Renaming all images with a consistent identifier (Spine_{modality}_{id}, so Spine_t1_0001.nii.gz, Spine_t2_0001.nii.gz etc)
2. Placing all volumes that belong to the same subject in the same folder (P001/Spine_t1.nii.gz, P001/Spine_t2.nii.gz)

Am I overlooking something?