Bad Segmentation for IHC Images with cytoplasm staining

[pouya-codes]

Hi,

I'm working on a research project to segment and classify IHC slides with membrane staining. I used your online tool (deepliif.org) to evaluate your model on my dataset, but it seems to have difficulty segmenting the cells correctly. I have attached a sample image along with the model's output for reference.

Are there any configurations or parameters I can modify to improve the results?

Thank you.

[sanadeem]

Thank you for your interest in our work. The current DeepLIIF model only supports nuclear markers (e.g. Ki67, ER, PR, TP53) and markers that express close to the boundary of the nuclei (e.g. CD3, CD8, ICOS), ref our Nature Machine Intelligence'22 paper. We will be releasing the DeepLIIF extension later this year for more complex markers such as PD-L1 (with membranous/cytosplasmic expression, as shown in your example).