I have WSI of Breast cancer and I want to run them through custom models and make heatmap

[lomshabhishek]

This repo looks like one of the most promising solutions to WSI processing.
Kindly help, Thanks.

[andreped]

Hello, @lomshabhishek!

There is a comprehensive tutorial on how to get started using pyFAST for processing WSIs in Python here.
It should contain the necessary details. If you have further questions, feel free to ask :]

If you just care about running your model on new WSIs and visualize the predictions on top of the WSI, I could recommend FastPathology.

[lomshabhishek]

I have got the segmentation layer on purple background, but I wanted to see the segmentation layer on original image.
While rendering it overlays on original image but when exported it uses purple background.

[andreped]

Yes, but the image you showed is all purple, indicating that there does not appear to be any detections. Given that you have a breast cancer model, the pipeline says that no cancer was detected. So there is likely something wrong. See my suggestions above for how to proceed.

When that is working then we can implement something to add the tumor predictions as overlay on top of the original WSI.

[lomshabhishek]

I have got the predictions, need these to masked over original image and save the image offline.

[andreped]

Great! In that case, the final step is using matplotlib. Hence, you have managed to do all relevant steps with pyFAST.

See my comment above for how to plot the WSI and prediction image side-by-side: #181 (reply in thread)

To plot the generated segmentation as overlay on top of the low-resolution WSI, see this thread.

Note that to plot segmentation on top of the low-res WSI, they likely need to be of equal image size. This can be achieved using OpenCV's resize function.

To save a figure in matplotlib, see savefig.

As the final steps are not FAST-related, I will let you try that on your own. Best of luck! :]

As you are now able to generate and view the heatmap using pyFAST, which was the original question, please accept my answer and resolve this Discussion thread.

If you have any future FAST-related questions or issues, please make a new Discussion.

[lomshabhishek]

Thanks a lot @andreped, you have been a great help :)

[andreped]

Thanks a lot @andreped, you have been a great help :)

Thanks, @lomshabhishek! :] Always fun to help!

If you have any other FAST-related issues, be sure to make a new Discussion.
I have dedicated some time to assist developers in using FAST for pathology applications, so don't be a stranger ;)

[lomshabhishek]

import fast

Download a nuclei segmentation model from the DataHub

model = fast.DataHub().download('nuclei-segmentation-model')

importer = fast.WholeSlideImageImporter.create(
image_path)

tissueSegmentation = fast.TissueSegmentation.create()
.connect(importer)

generator = fast.PatchGenerator.create(
256, 256,
magnification=20,
overlapPercent=0
).connect(importer)
.connect(1, tissueSegmentation)

segmentation = fast.SegmentationNetwork.create(
model.paths[0] + '/high_res_nuclei_unet.onnx',
scaleFactor=1./255.
).connect(generator)
stitcher = fast.PatchStitcher.create()
.connect(segmentation)
finished = fast.RunUntilFinished.create()
.connect(stitcher)

exporter = fast.TIFFImagePyramidExporter.create('segmented-nuclei-result_2.tiff')
.connect(importer)
.connect(finished)
.run()

import fast
import matplotlib.pyplot as plt
import numpy as np

Run importer and get data

imagePyramid = fast.TIFFImagePyramidImporter
.create('segmented-nuclei-result_2.tiff')
.runAndGetOutputData()

Extract specific level as FAST image

image = imagePyramid.getAccess(fast.ACCESS_READ).getLevelAsImage(level=2)

Convert FAST image to numpy ndarray and plot

image = np.asarray(image)

This should print: (2664, 3222, 3) uint8 255 0

print(image.shape, image.dtype, np.max(image), np.min(image))

plt.imshow(image)
plt.show()

[Peregalli]

Hi!
I have a similar problem. I am using the colon epithelium segmentation HE model and running with my WSI. I would like to get the predictions as a probability (or a heatmap) without output thresholding and save it as a tiff. I have been able to save a segmentation output as a .tiff file and get a segmentation image. But when I import .tiff file is already binary. I have seen that there is an attribute heatmapOutput in SegmentationNetwork class but when I set that parameter as True, I can not save the segmentation correctly. Another thing is that I am using getPatchAsImage insted of getLevelAsImage. I also run this code but the output shape is (14541, 8562, 1).

[andreped]

Hello, @Peregalli !

It would be great if you made a separate Q/A Discussion, as this topic is slightly different and this current Discussion has already been resolved.