Modelpublic

einarolafsson/cross-channel-nuclei-from-cellmask-cpsam

sourceHugging Facemitupdated 9d agoView on Hugging Face

0likes

Model Card

Cross-channel nuclei-from-cellmask (Cellpose-SAM)

Segments nuclei from the host cell mask channel alone — no nuclear stain required. A cross-channel model: it is given the cell image and predicts where the nuclei are, freeing the DAPI/Hoechst channel for another marker.

—Architecture: Cellpose-SAM (cpsam_v2)
—Model Zoo key: nuclei_from_cellmask_v1
—Checkpoint: nuclei_from_cellmask
—Trained by: einarolafsson

Use it in spaCR

This model is distributed through the spaCR Model Zoo. spaCR is an open-source package for spatial phenotype analysis of CRISPR screens and microscopy images.

bash

pip install spacr

Model Zoo (GUI)

Launch the GUI and open the Model Zoo:

bash

spacr

Find Cross-channel nuclei-from-cellmask (Cellpose-SAM) in the model list and press Download. The Model Zoo verifies the checkpoint's SHA-256 after download, so a truncated or substituted file is rejected rather than silently used.

Model Zoo (Python)

python

from spacr import model_zoo

entry = next(e for e in model_zoo.catalogue() if e.key == "nuclei_from_cellmask_v1")
path  = model_zoo.install(entry, dest="~/spacr_models")
print(path)   # verified local checkpoint

Mask generation

Point spaCR's mask generation at the downloaded checkpoint:

python

from spacr.core import preprocess_generate_masks

settings = {
    "src": "/path/to/images",
    "nucleus": "cellpose",
    "nucleus_model": str(path),     # the checkpoint fetched above
    "nucleus_diameter": 20,
}
preprocess_generate_masks(settings)

In the GUI the same thing is under Make masks — choose the downloaded model in the Cellpose model field for the relevant object.

API: :func:spacr.core.preprocess_generate_masks, :func:spacr.spacr_cellpose.generate_masks_from_imgs

Performance

model	train	train obj.	test	test obj.	CV	F1 @ IoU 0.5	AJI	Dice	final train loss	final val loss	val - train	best epoch
stock cpsam_v2 (no fine-tuning)	—	—	453	42487	—	0.2009	0.2856	0.4493	—	—	—	—
this model	—	not recorded	453	42487	no (single well-grouped split)	0.8881	0.7916	0.8774	—	—	—	best / 100

Scored on a well-grouped held-out split — no well appears in both train and test.

Per host cell line:

host	n	F1	AJI
HFF	157	0.9323	0.7462
HeLa	147	0.8596	0.8799
THP1	149	0.8610	0.7524

Objects are reference (ground-truth) objects. Training-set object counts and the per-epoch history were not preserved for this run, so the loss columns and training curves are unavailable.

Training data

Well-grouped split shared with the other cross-channel models, so no well leaks across train and test. 100 epochs from stock cpsam_v2, AdamW, lr 1e-5, weight decay 0.1.

Environment


cellpose (training)	4.2.1.1
cellpose (scoring)	4.0.9
GPU	NVIDIA GeForce RTX 3090 Ti
base weights	cpsam_v2

Files in this repository

path	what
`checkpoints/nuclei_from_cellmask`	data
`checkpoints/nuclei_from_cellmask_epoch_0010`	data
`checkpoints/nuclei_from_cellmask_epoch_0020`	data
`checkpoints/nuclei_from_cellmask_epoch_0030`	data
`checkpoints/nuclei_from_cellmask_epoch_0040`	data
`checkpoints/nuclei_from_cellmask_epoch_0050`	data
`checkpoints/nuclei_from_cellmask_epoch_0060`	data
`checkpoints/nuclei_from_cellmask_epoch_0070`	data
`checkpoints/nuclei_from_cellmask_epoch_0080`	data
`checkpoints/nuclei_from_cellmask_epoch_0090`	data
`qc/best_matched_ious.csv`	data
`qc/best_perimage.csv`	data
`qc/best_perimage_periou.csv`	data
`qc/best_periou.csv`	data
`qc/best_summary.json`	data
`qc/comparison_vs_stock.csv`	data
`qc/final_matched_ious.csv`	data
`qc/final_perimage.csv`	data
`qc/final_perimage_periou.csv`	data
`qc/final_periou.csv`	data
`qc/final_summary.json`	data
`qc/stock_cpsam_v2_matched_ious.csv`	data
`qc/stock_cpsam_v2_perimage.csv`	data
`qc/stock_cpsam_v2_perimage_periou.csv`	data
`qc/stock_cpsam_v2_periou.csv`	data
`qc/stock_cpsam_v2_summary.json`	data
`training/best_perimage.csv`	training metrics
`training/loss_per_epoch.csv`	training metrics
`training/metrics.csv`	training metrics
`training/nuclei_from_cellmask_training_curves.pdf`	training metrics
`training/nuclei_from_cellmask_training_curves.png`	training metrics
`training/report.json`	training metrics
`training/scores/best_matched_ious.csv`	training metrics
`training/scores/best_perimage.csv`	training metrics
`training/scores/best_perimage_periou.csv`	training metrics
`training/scores/best_periou.csv`	training metrics
`training/scores/comparison_vs_stock.csv`	training metrics
`training/scores/final_matched_ious.csv`	training metrics
`training/scores/final_perimage.csv`	training metrics
`training/scores/final_perimage_periou.csv`	training metrics
`training/scores/final_periou.csv`	training metrics
`training/scores/stock_cpsam_v2_matched_ious.csv`	training metrics
`training/scores/stock_cpsam_v2_perimage.csv`	training metrics
`training/scores/stock_cpsam_v2_perimage_periou.csv`	training metrics
`training/scores/stock_cpsam_v2_periou.csv`	training metrics
`weights/nuclei_from_cellmask`	data
`weights/nuclei_from_cellmask_best`	data

Limitations

—The held-out split is used for checkpoint selection, so it is validation data rather than a fully independent test set.
—Targets are automatic reference labels rather than hand-drawn ground truth.
—Predicts nuclei positions from cell morphology — expect degraded accuracy on unusual or highly confluent morphologies.