Where should your algorithm detect lymphocytes and plasma cells?

Dear participants,

As you have noticed, after each submission, we do some checks to make sure that your algorithm is running in the right way. We have been collecting some information and typical mistakes that we have found, which Mart and Witali have posted in this forum, we recommend you checking at least these three posts for some tips, and also to allow us to help you by making us editors of your algorithms (temporarily):

https://grand-challenge.org/forums/forum/tiger-601/topic/steps-before-submitting-to-l1-420/ https://grand-challenge.org/forums/forum/tiger-601/topic/please-add-one-the-organizers-to-your-algorithm-as-an-editor-or-user-512/ https://grand-challenge.org/forums/forum/tiger-601/topic/segmentation-artifacts-and-guidelines-for-tiger-example-algorithm-503/

But back to the topic of this post, where should your algorithm predict lymphocytes and plasma cells? We have noticed that some of you only submit predictions of cells in the tumor-associated stroma only. This is not the idea.

In L1, we evaluate the detection of lymphoctyes and plasma cells in every tissue compartment! This is also why, in the training set, you find those cells annotated also outside of the tumor-associated stroma. Therefore, if you have any post-processing step that removes cells that are not in the stroma, you do not need to do that, as this would penalize your score in the FROC analysis in L1!

Of course, when your algorithm goes to L2, you may want to exclude cells outside of the stroma if you score the TILs according to the recommendations of the TILs working group, but this would be only needed to compute a TILs score, and not needed for L1.

Regards, Francesco

EDIT: we discussed this during the webinar, now I also added this to the FAQ section.

Thank you for the clarification.

I was unclear that as I reading the detection rules in the evaluation page (https://tiger.grand-challenge.org/Evaluation/).

Participants will have to submit a list of (x,y) coordinates with the location of each predicted TIL (see Submission requirements for details), which will be compared with the manual reference standard.

why don't you change the explanation in detection rule? As it could be interpreted as to submit the lymphoplasma cells in stroma area only if we follow the definsition of TIL from knowledge.

Thanks for your suggestion, I just updated the text on the page as follows:

"Participants will have to submit a list of (x,y) coordinates with the location of each predicted cell (see Submission requirements for details), which will be compared with the manual reference standard. Note that in Leaderboard 1, we expect algorithms to predict the location of cells (lymphocytes and plasma cells) in any tissue compartments, not solely limited to the tumor-associated stroma. This is because, in this phase, we only assess the "computer vision performance" of algorithms."

I hope this is more clear and avoid confusion, we thought it was clear from the description, the discussion we had during the webinar, and the fact that, in the training set, lymphocytes and plasma cells are annotated also outside of the tumor-associated stroma, but it is good to clarify this.

Regards, Francesco