DWI/ADC lesion delineation mapping

In the description for annotation there's this line:

"for every annotation in this folder, all lesion delineations (if any) will always clearly map to observations in DWI/ADC imaging."

how exactly can we do this mapping? In most cases the annotation is for the T2W image and I don't know how to map it to DWI/ADC.

Sorry if this was covered somewhere else.

Hi Yizhou,

When we say that "all lesion delineations (if any) will always clearly map to observations in DWI/ADC imaging" in picai_labels/csPCa_lesion_delineations/human_expert/, we mean that if you open these images (e.g. using ITK-SNAP), you will notice that the annotations share the same spatial location as the lesions that they refer to. Or in other words, the annotations are registered to their corresponding DWI/ADC images. You will notice that this statement is true, even when the annotations and DWI/ADC images share different spatial dimensions and resolutions. Here is an example case (patient 10522, study 100532 from the Public Training and Development Dataset), visualized using ITK-SNAP.

However, please note, that while all human expert-derived annotations may be registered to their corresponding DWI/ADC images, those DWI/ADC images themselves aren't necessarily registered to the T2W images for that same case. This issue only persists across a fraction of the training datasets. For the Hidden Validation and Tuning Cohort and the Hidden Testing Cohort, this won't be an issue as we have manually registered all images for the same case.

If you're struggling with where/how to get started due to all the different nuances of the dataset + tasks of this challenge, I would strongly recommend that you first reproduce one of the public baseline models that we've provided (which covers everything from data preprocessing, model training and algorithm submission to the leaderboard). And then adapt that setup to your needs, strategy and objectives.

Hope this helps.