zizhaozhang / nmi-wsi-diagnosis Goto Github PK
View Code? Open in Web Editor NEWPathologist-level interpretable whole-slide cancer diagnosis with deep learning
Home Page: https://www.nature.com/articles/s42256-019-0052-1
Pathologist-level interpretable whole-slide cancer diagnosis with deep learning
Home Page: https://www.nature.com/articles/s42256-019-0052-1
dear sir, I have download the data set from https://docs.google.com/forms/d/e/1FAIpQLSfaBugmwdR_BalbwrhkHK64xysaXjiE_JLjOKQVlci2DVTvNg/viewform?usp=sf_link
I find that in the the folder “slide”,in val_test_diagnosis.json,you mentioned picture like“1153831,...,118633”, but in "Slide\Img", I couldn't find picture like“1153831,...,118633”,Is the data set not uploaded completely? thank you
You have done great job! But i am confused about some code.
`def data_loader(path, batch_size, imSize,
mean=[0.5,0.5,0.5], std=[0.5,0.5,0.5],
ignore_val=44, pos_val=255, neg_val=155, pos_class=[0,1], neg_class=[2]):
# pos_class and neg_class in the folder name for keras ImageDataGenerator input
# 0,1,2 are low, high, normal
def imerge(img_gen, mask_gen):
for (imgs, img_labels), (mask, mask_labels) in itertools.zip_longest(img_gen, mask_gen):
# compute weight to ignore particular pixels
# mask = np.expand_dims(mask[:,:,:,0], axis=3)
mask = mask[:,:,:,0]
weight = np.ones(mask.shape, np.float32)
weight[mask==ignore_val] = 0.5 # this is set by experience
# In mask, ignored pixel has value ignore_val.
# The weight of these pixel is set to zero, so they do not contribute to loss
# The returned mask is still binary.
# compute per sample
for c, mask_label in enumerate(mask_labels):
assert(mask_labels[c] == img_labels[c])
mask_pointer = mask[c]
if mask_label in pos_class:
assert(np.where(mask_pointer == neg_val)[0].size == 0)
mask_pointer[mask_pointer==pos_val] = 1
elif mask_label in neg_class:
assert(np.where(mask_pointer == pos_val)[0].size == 0)
mask_pointer[mask_pointer==neg_val] = 0
else:
print ('WARNING: mask beyond the expected class range')
mask_pointer /= 255.0
mask_pointer[mask_pointer==ignore_val] = 0
assert set(np.unique(mask)).issubset([0, 1])
# assert set(np.unique(weight)).issubset([0, 1])
# img, mask = preprocess(imgs, mean, std, mask)
yield imgs, mask, weight, img_labels
`
code above in segmentation/data_gen.py. what is the aim to calculate 'mask_pointer'? it seems that it is not returned finally?
And 'assert set(np.unique(mask)).issubset([0, 1])' is apparently cannot be satisfied since the pixels in mask only belong to [44 255 155].
The "code availability" section in your paper refers to this repo. Unfortunately, it is still empty.
I am sure many readers will, like me, be keen to have a look at the implementation. Thanks!
Hi, I was wondering where was this file generated?
Dear author,
About the III-Report dataset, are there any masks to indicate the tumor area in each image? So I can compare the tumor detection (heat maps) with the ground truth region.
Or, can we know their coordinates in the original whole slide images?
Great work. I want to run your code on my WSI dataset. But I do not have annotation of those imaging data. Whole dataset is divided into group1 and group2. Would it be possible to run code without region of interest ? In that case, one can take all information from the slides without defining specific area ?
Looking forward to hear from you.
Thanks in advance
Dear authors,
Thank you for sharing your work.
Your paper mentioned that there are 620 slides for training, 193+100 for validation and testing. These match the 620+293 entries in the supplied {train/val_test}_diagnosis.json.
On the other hand, there are only 482 downloadable slides. It is close to 336+142=478 entries in the supplied {train/val_test}_diagnosis_partial.json (used by anno_parser/gen_patches.py).
My questions are:
Thank you!
Mr Zhang, I would like to take the liberty to disturb you. Recently, we encountered a problem when we reproduced the code of the thesis pathologist level interpretable whole slide cancer diagnosis with deep learning GitHub published on nature. We downloaded the thesis data according to your requirements in data preparation and established the corresponding directory structure. When we reproduced the first network S-Net, we found that, Although there is an error, it still runs successfully after modification and generates the weight file. However, when we reproduce the second network DNET, we find that there are many directory structures not mentioned in the SH script and the code of Tai. These problems will lead to the lack of a parameter in the gas. The third network a-nt also has class annotation G! Many of them are how many 7, which are similar problems in the code. These problems make it very difficult for us to reproduce your paper, because we can't find out whether Dai didn't mention these directories or the directory structure of the code. In order to better learn the content of your article, we want to know the whole content when you write the code, which reflects these directories, but there is no explanation? Can you give the directory structure, if it is convenient for you, we also want to know where the specific path needs to be modified? What is the most specific directory name of your code operation? Sh full name is the end result. Thank you very much. What is train? Is it a model or the name of a code file? Dear Dr Zhang's comments [tg2j: which of these problems (refers to the directory structure without hope this letter finds you well!)? What is fags? File? Code? Parameter? Recently, we are reproducing the models proposed in this paper: pathologist level, The hope that you / the other authors can kindly help us with some problems we accounted for The instructions in data preparation the first S-Net was successfully reproduced and the weight file is generated fine.sh and the train file. This problem leads tosomewhat a missing parameter in FLAGS. We also find the similar problem with the thirda-netStruggling with the structure of code directory, we found it hard to reproduce the modelproposed in your paper. So, we want to understand the overall directory structure of thecode. It would also be great for you to show us where the path needs to be modified, andthe final result of the code if it is convenient to let us knowThank you very much for your time and patience. Looking forward to hearing from you soon
Dear authors, I want download the datasets of your experiment. But I can't download the datasets with your download link successfully. The error is "Too many requests.Sorry, this link has been automatically turned off for now. Learn more about traffic limits.". Could you please provide a new download link? Thank you very much!
A declarative, efficient, and flexible JavaScript library for building user interfaces.
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
An Open Source Machine Learning Framework for Everyone
The Web framework for perfectionists with deadlines.
A PHP framework for web artisans
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
Some thing interesting about web. New door for the world.
A server is a program made to process requests and deliver data to clients.
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
Some thing interesting about visualization, use data art
Some thing interesting about game, make everyone happy.
We are working to build community through open source technology. NB: members must have two-factor auth.
Open source projects and samples from Microsoft.
Google ❤️ Open Source for everyone.
Alibaba Open Source for everyone
Data-Driven Documents codes.
China tencent open source team.