iSight-target: target-cell selection for iSight-cell

Given all segmented cells of an IHC image and the image's tissue × cell-type class, iSight-target decides which cells are the cells of interest (tumour cells in a carcinoma, hepatocytes in liver, and so on). It has one binary head per class in meta/classes_43.csv — 43 classes — on a UNI2-h backbone. iSight-cell scores the cells this model selects.

Files

config.json                          model configuration (fetch this with the checkpoint)
checkpoints/iSight-target.pt     weights (model state dict only)
meta/classes_43.csv                  the 43 classes: cls43, head_idx, tissue, cell_type

from huggingface_hub import hf_hub_download
cfg  = hf_hub_download("zhihuanglab/iSight-target", "config.json")
ckpt = hf_hub_download("zhihuanglab/iSight-target", "checkpoints/iSight-target.pt")

Links

• Code: github.com/zhihuanglab/iSight (isight_cell/target_cell/, isight_cell/code/pipeline/select_target_v2.py)
• Cell-level model: zhihuanglab/iSight-cell

License

PENN Academic Software License Agreement: non-commercial research use only.

Contact

Zhi Huang — zhi.huang@pennmedicine.upenn.edu