iTranslated by AI
[Image Analysis] Cell Segmentation from Micrographs with Cellpose Part 2: Batch Processing
In the previous article, I introduced how to use the Cellpose GUI and how to create custom models.
However, since the Cellpose GUI does not support batch processing, you cannot repeat the process on other images even if you have developed a versatile model.
In this article, I will introduce examples of applying Cellpose models to multiple images using Python command-line processing.
We will use the cellpose virtual environment created last time. Start the virtual environment from the Anaconda Prompt.
conda activate cellpose
There are two ways to proceed: processing directly from the command line or starting and running Jupyter Notebook/Lab.
【 Execution from the Command Line 】
Use the python -m cellpose command.
There are many arguments, but the minimum ones needed are listed below:
-
--dir:
Specify the path to the folder containing the images. -
--pretrained_model:
Specify the model to be used for prediction. Provide the model name if using one provided by Cellpose, or the path to the model file if using one you created yourself. -
--diameter:
The pixel diameter of the structures/objects you want to detect. If nothing is specified, 30 is used. Specifying--diameter 0will use the value saved in the model. If using a model provided by Cellpose, it might estimate it for each image?? -
--chan/--chan2:
Channel numbers to use. The default is 0, assuming grayscale. Depending on the model, you may need to specify the cytoplasm and nuclei channels. -
--savedir:
The destination for saving predicted label images. If not specified, the results will be output to the folder specified by--dir. *Note: However, the Numpy data file for the predicted labels, "_seg.npy", will be saved in the--dirfolder. -
--save_png/--save_tif:
Instructs whether to save the predicted label images as png or tif in addition to the "_seg.npy" file. The save destination is the one specified by--savedir. If not specified, they are saved in the--dirpath. -
--save_rois:
A zip file containing ROI outline coordinate information. This is the same as what is obtained viaFile > Save outlines as .zip archive of ROI files for ImageJin the Cellpose GUI and can be handled from the ImageJ ROI Manager. -
--use_gpu:
Instructs to use the GPU. Include this if a GPU-enabled Cellpose environment is available.
help
usage: __main__.py [-h] [--version] [--verbose] [--use_gpu] [--gpu_device GPU_DEVICE] [--check_mkl] [--dir DIR]
[--image_path IMAGE_PATH] [--look_one_level_down] [--img_filter IMG_FILTER]
[--channel_axis CHANNEL_AXIS] [--z_axis Z_AXIS] [--chan CHAN] [--chan2 CHAN2] [--invert]
[--all_channels] [--pretrained_model PRETRAINED_MODEL] [--add_model ADD_MODEL] [--unet]
[--nclasses NCLASSES] [--no_resample] [--net_avg] [--no_interp] [--no_norm] [--do_3D]
[--diameter DIAMETER] [--stitch_threshold STITCH_THRESHOLD] [--min_size MIN_SIZE] [--fast_mode]
[--flow_threshold FLOW_THRESHOLD] [--cellprob_threshold CELLPROB_THRESHOLD]
[--anisotropy ANISOTROPY] [--exclude_on_edges] [--augment] [--save_png] [--save_tif] [--no_npy]
[--savedir SAVEDIR] [--dir_above] [--in_folders] [--save_flows] [--save_outlines] [--save_rois]
[--save_ncolor] [--save_txt] [--train] [--train_size] [--test_dir TEST_DIR]
[--mask_filter MASK_FILTER] [--diam_mean DIAM_MEAN] [--learning_rate LEARNING_RATE]
[--weight_decay WEIGHT_DECAY] [--n_epochs N_EPOCHS] [--batch_size BATCH_SIZE]
[--min_train_masks MIN_TRAIN_MASKS] [--residual_on RESIDUAL_ON] [--style_on STYLE_ON]
[--concatenation CONCATENATION] [--save_every SAVE_EVERY] [--save_each]
[--model_name_out MODEL_NAME_OUT]
Cellpose Command Line Parameters
optional arguments:
-h, --help show this help message and exit
--version show cellpose version info
--verbose show information about running and settings and save to log
Hardware Arguments:
--use_gpu use gpu if torch with cuda installed
--gpu_device GPU_DEVICE
which gpu device to use, use an integer for torch, or mps for M1
--check_mkl check if mkl working
Input Image Arguments:
--dir DIR folder containing data to run or train on.
--image_path IMAGE_PATH
if given and --dir not given, run on single image instead of folder (cannot train with this
option)
--look_one_level_down
run processing on all subdirectories of current folder
--img_filter IMG_FILTER
end string for images to run on
--channel_axis CHANNEL_AXIS
axis of image which corresponds to image channels
--z_axis Z_AXIS axis of image which corresponds to Z dimension
--chan CHAN channel to segment; 0: GRAY, 1: RED, 2: GREEN, 3: BLUE. Default: 0
--chan2 CHAN2 nuclear channel (if cyto, optional); 0: NONE, 1: RED, 2: GREEN, 3: BLUE. Default: 0
--invert invert grayscale channel
--all_channels use all channels in image if using own model and images with special channels
Model Arguments:
--pretrained_model PRETRAINED_MODEL
model to use for running or starting training
--add_model ADD_MODEL
model path to copy model to hidden .cellpose folder for using in GUI/CLI
--unet run standard unet instead of cellpose flow output
--nclasses NCLASSES if running unet, choose 2 or 3; cellpose always uses 3
Algorithm Arguments:
--no_resample disable dynamics on full image (makes algorithm faster for images with large diameters)
--net_avg run 4 networks instead of 1 and average results
--no_interp do not interpolate when running dynamics (was default)
--no_norm do not normalize images (normalize=False)
--do_3D process images as 3D stacks of images (nplanes x nchan x Ly x Lx
--diameter DIAMETER cell diameter, if 0 will use the diameter of the training labels used in the model, or with
built-in model will estimate diameter for each image
--stitch_threshold STITCH_THRESHOLD
compute masks in 2D then stitch together masks with IoU>0.9 across planes
--min_size MIN_SIZE minimum number of pixels per mask, can turn off with -1
--fast_mode now equivalent to --no_resample; make code run faster by turning off resampling
--flow_threshold FLOW_THRESHOLD
flow error threshold, 0 turns off this optional QC step. Default: 0.4
--cellprob_threshold CELLPROB_THRESHOLD
cellprob threshold, default is 0, decrease to find more and larger masks
--anisotropy ANISOTROPY
anisotropy of volume in 3D
--exclude_on_edges discard masks which touch edges of image
--augment tiles image with overlapping tiles and flips overlapped regions to augment
Output Arguments:
--save_png save masks as png and outlines as text file for ImageJ
--save_tif save masks as tif and outlines as text file for ImageJ
--no_npy suppress saving of npy
--savedir SAVEDIR folder to which segmentation results will be saved (defaults to input image directory)
--dir_above save output folders adjacent to image folder instead of inside it (off by default)
--in_folders flag to save output in folders (off by default)
--save_flows whether or not to save RGB images of flows when masks are saved (disabled by default)
--save_outlines whether or not to save RGB outline images when masks are saved (disabled by default)
--save_rois whether or not to save ImageJ compatible ROI archive (disabled by default)
--save_ncolor whether or not to save minimal "n-color" masks (disabled by default
--save_txt flag to enable txt outlines for ImageJ (disabled by default)
Training Arguments:
--train train network using images in dir
--train_size train size network at end of training
--test_dir TEST_DIR folder containing test data (optional)
--mask_filter MASK_FILTER
end string for masks to run on. use "_seg.npy" for manual annotations from the GUI. Default:
_masks
--diam_mean DIAM_MEAN
mean diameter to resize cells to during training -- if starting from pretrained models it
cannot be changed from 30.0
--learning_rate LEARNING_RATE
learning rate. Default: 0.2
--weight_decay WEIGHT_DECAY
weight decay. Default: 1e-05
--n_epochs N_EPOCHS number of epochs. Default: 500
--batch_size BATCH_SIZE
batch size. Default: 8
--min_train_masks MIN_TRAIN_MASKS
minimum number of masks a training image must have to be used. Default: 5
--residual_on RESIDUAL_ON
use residual connections
--style_on STYLE_ON use style vector
--concatenation CONCATENATION
concatenate downsampled layers with upsampled layers (off by default which means they are
added)
--save_every SAVE_EVERY
number of epochs to skip between saves. Default: 100
--save_each save the model under a different filename per --save_every epoch for later comparsion
--model_name_out MODEL_NAME_OUT
Name of model to save as, defaults to name describing model architecture. Model is saved in
the folder specified by --dir in models subfolder.
1. Preparing the Model
Continuing from last time, we will use brightfield images from a phase-contrast microscope and use the custom-made model. In this article, it is assumed that the model is placed on the desktop as "models/model_filename".
If you use a pretrained model, specifying the model name will download the model file for you.
2. Preparing the Prediction Image Folder
First, prepare a folder containing the images you want to predict. For clarity, the folder is placed on the desktop.
3. Running the Prediction
Run the following command. Specify the path to the folder containing the images for prediction in the --dir argument, and the path to the model file in the --pretrained_model argument. (It seems that absolute paths are preferred.)
python -m cellpose --dir "C:\Users\Ryota_Chijimatsu\Desktop\Prediction" --pretrained_model C:\Users\Ryota_Chijimatsu\Desktop\models\CP_20230728_122230 --chan 0 --chan2 0 --diameter 0 --save_png --save_rois
Execution screen
4. Checking the Results
Once finished, you will get output like the following.
Files ending in "_cp_masks.png" are label images that, when opened in ImageJ, have different intensities for each detected cell.
Files ending in "_cp_output.png" show summary images of the predictions side-by-side.
"_rois.zip" files contain ROI information to be used in ImageJ. (Open the original image in ImageJ -> Analyze > Tools > ROI Manager -> Open the Roi zip file from More > Open in the ROI Manager.)
【 Execution from a Notebook 】
https://cellpose.readthedocs.io/en/latest/notebook.html
https://nbviewer.org/github/MouseLand/cellpose/blob/master/notebooks/run_cellpose.ipynb
This is also introduced in this book, so please refer to it.
First, start Jupyter Notebook or Jupyter Lab.
jupyter lab
1. Loading Libraries
The models module within cellpose is used for loading prediction models and executing predictions.
The io module is used for reading images for prediction and writing out prediction results.
from cellpose import models, io
2. Loading the Model
Load the model using the .CellposeModel() method. When using a model provided by cellpose, specify it like model_type='cyto'.
When using a custom model, specify the path to the model file using the pretrained_model= argument.
model_path = r"C:\Users\Ryota_Chijimatsu\Desktop\models\CP_20230728_122230"
model = models.CellposeModel(pretrained_model=model_path)
3. Loading Images for Prediction
Load the images to be used for prediction. We use cellpose's io.imread, but other tools can be used as long as they provide image data as a NumPy array.
For prediction, use either a single image or a list of multiple images.
img_path = r"C:\Users\Ryota_Chijimatsu\Desktop\Prediction\BF_1_XY01_00001_CH1.tif"
img = io.imread(filename=img_path)
from glob import glob
# Create a list of image file paths
img_list = glob(r"C:\Users\Ryota_Chijimatsu\Desktop\Prediction\*.tif")
# Convert to a list of images
img_list = [ io.imread(img) for img in img_list]
4. Executing Predictions
Use the .eval() method. Arguments include a single image or a list of images, and parameters like diameter= and channels=.
Setting diameter=None allows the model to estimate the object diameter automatically.
In the channels= argument, specify a list such as [channel1_number, channel2_number]. The number 0 represents grayscale or None. This varies depending on the model, so you need to know which channels the model was trained on.
Let's execute the prediction and check its output.
result1 = model.eval(x=img,
diameter=None,
channels=[0,0])
result2 = model.eval(x=img_list,
diameter=None,
channels=[0,0])
5. Checking the Results
The results are returned as multiple values.
Among them, the first return value is the predicted label image.
plt.imshow(result1[0])
plt.colorbar()
When predicting with an image list, the first return value is also a list of label images.
fig, axs = plt.subplots(1,3, figsize=(10,10))
axs[0].imshow(result2[0][0])
axs[1].imshow(result2[0][1])
axs[2].imshow(result2[0][2])
Summary of Prediction Results
If necessary, you can use the prediction result information to check a summary image as follows.
from cellpose import plot
import matplotlib.pyplot as plt # pip install matplotlib
fig = plt.figure(figsize=(12,5))
plot.show_segmentation(fig, img=img, maski=result1[0], flowi=result1[1][0])
plt.tight_layout()
plt.show()
6. Exporting the Results
Use the io.masks_flows_to_seg() function to save the "_seg.npy" file locally.
io.masks_flows_to_seg(images=img, masks=result1[0], flows=result1[1], diams=None, file_names="tmp")
Use the io.save_to_png() function to export the predicted label images.
io.save_to_png(images=img, masks=result1[0], flows=result1[1], file_names="label.png")
【 Cellpose in QuPath 】
A Cellpose extension for the free image analysis software QuPath has been released.
Details are introduced in my book.
Not only can you perform predictions on multiple images/regions using published Cellpose models, but you can also use models created with the Cellpose GUI in QuPath.
Furthermore, it is also possible to create prediction models within QuPath. QuPath has more annotation tools than the Cellpose GUI, making labeling easier.
Even for large data such as whole slide images, you can train models by labeling arbitrary areas within the image.
Example of ground truth labels
While grayscale images are used when applying Cellpose to brightfield images, you can also perform color deconvolution in QuPath to use Hematoxylin or DAB channels with Cellpose.
Discussion