Tile grid shifts upon zooming

[Punzo]

From Steve: as Bill reported there's a shift to the down-right when high res tiles come in which may be a problem with the coordinate systems or maybe the data but probably the viewer.

[Punzo]

we need to test, it looks like the images get a shift, so it is not a panning of the camera.

Steps to reproduce: go to https://dev-slim-viewer.canceridc.dev/studies/1.3.6.1.4.1.5962.99.1.2440258434.818002713.1625937929090.3.0/series/1.3.6.1.4.1.5962.99.1.2440258434.818002713.1625937929090.2.0

zoom at the maximum level and just wait. When the high resolution tiles get loaded, there is a visible shift, between the low resolution and the high resolution images.

[Punzo]

@hackermd maybe a bug in static method deriveImageGeometry? https://github.com/MGHComputationalPathology/dicom-microscopy-viewer/blob/3b6b5afaae419e2f006e90fba297ad9612de6a8a/src/channel.js#L451

[Punzo]

I perfomed further testing on this. In the following video we can see that the "image shift" occurs only when the tile do a transition between "preview" (i.e. when is downloading) to fully loaded:

https://www.dropbox.com/s/m9ccicz5tstnr1x/2021-07-26%2015-29-26.mp4?dl=0

Therefore the fully downloaded images position is right in world coordinates (in the video you can focus on the annotation and see that the relative distances are maintained when zooming).

I agree seeing this shift of the images during the download transition is annoying, but it looks like it is not related to the grid parametes defined in deriveImageGeometry method (e.g. the offset: https://github.com/MGHComputationalPathology/dicom-microscopy-viewer/blob/3b6b5afaae419e2f006e90fba297ad9612de6a8a/src/channel.js#L589), but it is internal in OpenLayer.

@pieper @fedorov @hackermd how much priority has this issue? Should I try to further debug/investigate this issue in OpenLayer (i.e. trying to understand what happens to the tile world coordinates during the download)?

[pieper]

Thanks for investigating further @Punzo. We'll need to discuss what priority this has. So far it's been stated by @dclunie that this wouldn't be acceptable for clinical pathology review systems. But IDC may be okay with a viewer that is "good enough" to see the images. At the same time we don't want people to think that the data itself is somehow wrong or that the dicom conversion of the data has introduced issues. It also depends on how much effort fixing it is likely to require.

[Punzo]

Thanks for investigating further @Punzo. We'll need to discuss what priority this has. So far it's been stated by @dclunie that this wouldn't be acceptable for clinical pathology review systems. But IDC may be okay with a viewer that is "good enough" to see the images. At the same time we don't want people to think that the data itself is somehow wrong or that the dicom conversion of the data has introduced issues. It also depends on how much effort fixing it is likely to require.

ok then for the moment I will stop working on it until we have the occasion to discuss about it.

[fedorov]

Would also be good to hear from @ahomeyer about how important this issue is for the researchers.

[dclunie]

don't stop working on this ... there is clearly a problem that needs to be fixed, regardless of the user ... it's not annoying, it's unprofessional

[Punzo]

don't stop working on this ... there is clearly a problem that needs to be fixed, regardless of the user ... it's not annoying, it's unprofessional

we will definitely come around with a proper fix. However, as any bug involving third party libraries, it may take time and resources. It is @pieper @fedorov call to assign the level of priority for the issues and allocate resources. I will look forward to discuss it in the a videocall at the OHIF-IDC prioritization meetings (as I largely anticipated I will be in vacation in August, so for me it would be in September).

[ahomeyer]

I cannot access the dev-slim-viewer.canceridc.dev link above but took a look at the video.

We had a similar issue in the past with Hamamatsu NDPI files. It turned out, that higher levels in the image pyramid were shifted to some small degree and did not correspond to proper downsampled versions of the base layer. We had to solve this problem by resampling the entire image pyramid from the base layer.

For research applications, I don't find this shift too disruptive. Also, for the pathomics use case, we only consider full resolution base level images. It should be ensured, however, that there is no fundamental problem with image conversion/loading.