This notebook tests the cellpose algorithm to segment cells. We found that the model (cyto3) worked better after a preprocessing step were we apply a gaussian filter to each channel and then rescale the intensity.
# Test cellpose to segment cells
This notebook tests the cellpose algorithm to segment cells. We found that the model (cyto3) worked better after a preprocessing step were we apply a gaussian filter to each channel and then rescale the intensity.
```{python}
from cellpose import models
from glob import glob
import imageio.v3 as iio
import skimage as ski
import numpy as np
import matplotlib.pyplot as plt
```
```{python}
model = models.CellposeModel(gpu=True, model_type='cyto3')
```
```{python}
ims_to_cellpose = []
ims_cyto = []
ims_cyto_pre = []
for im_fn in glob('data/ome-tiff/*/*tif')[:10]:
im = iio.imread(im_fn)
# maximum intensity projection
im_nuclei = im[0].max(axis=0)
im_cytoplasm = im[1].max(axis=0)
im_cyto_pre = im_cytoplasm
# im_cyto_pre = ski.exposure.rescale_intensity(np.clip(im_cytoplasm,0,30), in_range=(0,30))
# im_cyto_pre = ski.exposure.rescale_intensity(np.clip(im_cytoplasm, 0, 20), out_range=(0.0,1.0))
im_cyto_pre = ski.filters.gaussian(im_cyto_pre, sigma=0.5)
im_cyto_pre = ski.exposure.rescale_intensity(im_cyto_pre,out_range=(0,1))
im_nuclei_pre = im_nuclei
im_nuclei_pre = ski.filters.gaussian(im_nuclei_pre, sigma=1)
im_nuclei_pre = ski.exposure.rescale_intensity(im_nuclei_pre,out_range=(0,1))
ims_to_cellpose.append(np.stack([im_cyto_pre, im_nuclei_pre], axis=0))
ims_cyto.append(im_cytoplasm)
ims_cyto_pre.append(im_cyto_pre)
# break
```
```{python}
# fig, axs = plt.subplots(len(ims_cyto), 2, constrained_layout=True, figsize=(6, 3*len(ims_cyto)))
# for idx in range(len(ims_cyto)):
# axs[idx,0].imshow(ims_cyto[idx], cmap='gray')
# axs[idx,1].imshow(ims_cyto_pre[idx], vmin=0, vmax=1, cmap='gray')
# axs[idx,0].axis('off')
# axs[idx,1].axis('off')
```
```{python}
masks, flows, styles = model.eval(ims_to_cellpose, channels=[0,1], diameter=100, normalize=True, min_size=30)
```
```{python}
fig, axs = plt.subplots(len(masks), 3, constrained_layout=True, figsize=(9, 3*len(masks)))
for idx, mask in enumerate(masks):
axs[idx, 0].imshow(ims_cyto[idx], cmap='magma')
axs[idx, 1].imshow(ims_cyto_pre[idx], cmap='grey')
axs[idx, 2].imshow(ims_cyto_pre[idx], cmap='grey')
axs[idx, 2].imshow(ski.color.label2rgb(mask), alpha=0.5)
axs[idx, 2].contour(mask, linewidths=0.1, colors='red')
axs[idx,0].axis('off')
axs[idx,1].axis('off')
axs[idx,2].axis('off')
```