adalbertocq / histomorphological-phenotype-learning Goto Github PK

Hello @AdalbertoCq,
My intention is to find tile vector representations for a given single svs file.
If I understand corrently, the real_hdf5 is the h5 file we get as output from DeepPath preprocess, and the checkpoint is weight we get from step 1 (I used your provided weights here).
When I try to run the script run_representationspathology_projection.py, it gives following error:

/models/selfsupervised/BarlowTwins.py", line 85, in __init__
self.num_samples = data.training.images.shape[0]
AttributeError: 'NoneType' object has no attribute 'images'

This error esstially happens because incorrect setting of dataset argument when I chase through the code if I'm right.

Can you explain more about what is the argument dataset of run_representationspathology_projection.py? Which dataset it is refering? And more importantly, why do I need another dataset during vectorising tiled images, given I already have the image h5 file (real_hdf5) and a pre trained model (checkpoint .ckt file).

Thank you in advance

Hi!
Thank you for your wonderful work!

I'm trying to load your trained networks, but I'm having some difficulties.
I managed to load the graph correctly, but I can't seem to find which tensors correspond to input/outputs... There are some gradients and Adam tensors and operations everywhere.

Would you think it would be possible for you to propose simpler weights/models if we are only looking to use it for inference?
Or, alternatively, a small script/notebook that shows how to load and use the model (Not inside your particular pipeline, but on its own.)

Thank you!

Hello,@AdalbertoCq
I am trying run through the pipeline of Mapping an external cohort to existing clusters. For simplicity I am using just one WSI image as the external cohort.
On Step 4 including metadata in h5, I notice a csv file contains "luad", "os_event_ind", "os_event_data" column. I cannot create my own csv file, Where are these data come from? I secrch around and find nothing.

Might a silly question :)

Thank you

Hi, @AdalbertoCq ,

Thanks for sharing a wonderful work！

I ran into some statistical problems about Cox proportional hazard regression while reading your work.

You mentioned in the article that you implement a central logarithmic transformation after using multiplicative replacement for WSI vector representations to ensure independence between the covariates. But I noticed that after these transformations, the sum of the features in each row is almost equal to 0. Whether this is a hidden linear relationship and affects the calculation of Cox covariates, such as making the value of a covariate tend to infinity？

I would be grateful if you could answer these questions to me.

Thanks.

Hi, very nice work!

I want to extract the feature vectors for some of my H&E data, on tiles directly at 20x magnification.
However, I have seen that the pre-trained models have been trained on a 5x downscaled version of the images.

Have you ever tested the feature encoding for images with no downscaling?
Is it going to cause problems?

Thank you very much in advance