BnT Cells Unsupported

[DarioS]

It is typical for imaging mass cytometry to have a cell category named BnT. Why does this algorithm not support such an ambiguity?

Previous IMC publications detail so-called ‘BnT’ cells, interacting B and T cells that could not be separated by segmentation

Due to the poor 1 µm per pixel resolution, it is often beyond the capability of the instrument to distinguish such cell types. Nonetheless, DeepCell Types makes confident claims about a cell in an image being a particular cell type when it is fundamentally impossible to do so.

[LukasHats]

It is not fundamentally impossible to do that. If you use clustering for annotation, that might be the case. But there are other techniques. For example visual thresholding and subsequent scoring like scimap offers. If cell segmentation is good, we can still assume a generally higher expression of CD20 on the B cell side and CD3 on the T cell side even with heavy spillover, which is detectable with a manual thresholding and scoring approach, but probably impossible with clustering. Sure, this will not resolve all instances, but on average, this can be assumed, given a good segmentation (e.g. with the new cellpose or cellsam).