ucsd-ai4h / covid-ct Goto Github PK

How the non-covid images are collected. ? I did not find any information in Data creation step of the paper

Hello, when i run the code 'CT_predict', a error occurred: alpha doesn't exist and i found this txt also doesn't exist.
f = open('COVID-CT/model_result/DenseNet_{}.txt'.format(alpha), 'a+')

Thanks for providing images again!

Dears

Based on your implementation is it possible to show any visual results? Something like a heatmap...

pls list all the requirement(packages)..

Thank you for your effort. Please I have two questions.

1. What is the required total number of epochs to get the reported results? Because in DenseNet_predict.py file I noticed that there are 3000 epochs which is extremely high!!
2. In the DenseNet_predict.py, did you tuned the networks pretrained weights or you are using them as is?
   Thanks

i cloned, then i install everything, but when i try to run "python CT_predict.py" i am getting following error: No such file or directory: 'COVID-CT/train_set'" can you please rell, what chnges i need to made after clone.

yes thanks for providing ct images.
Thanks

Thank you very much for providing the dataset and baseline model! I am participating in the COVID-19 CT competition (https://covid-ct.grand-challenge.org/), but I cannot find the format of the submission file on the competition page. Could you provide a sample submission file or describe the format of the submission file in detail?

There's a new dataset that HM Hospitales in Spain is sharing with qualified researchers, consisting of anonymized medical records from COVID-19 patients:

https://www.hmhospitales.com/prensa/notas-de-prensa/comunicado-covid-data-save-lives

From April, 25th onwards, the 'COVID DATA SAVE LIVES' contents will be distributed online, free of cost and other access barriers to worldwide health care institutions, universities, and scientific organizations...

To access this initiative, it is mandatory to submit an application via e-mail to [email protected] which will be assessed by the HM Hospitales Data Science Commission and, where appropriate, reviewed by the HM Hospitales Clinical Research Ethics Committee.

The resolution of photos is not the same, but also contains a large number of annotations

Hello,

Thank you for your data set .It is very valuable.

I am a graduate student, I clone your zip .But there is bug in code.
So I want a new updated code. My email [email protected]

Thank you again.

Excuse me~ when I run the baseline methods, I met the following question:
RuntimeError: Given groups=1, weight of size 32 3 3 3, expected input[10, 1, 225, 225] to have 3 channels, but got 1 channels instead.

I have no idea, can you help me?

When I run DenseNet_predict.py, the error comes "No such file or directory: 'new_data/newtxt/train.txt'."

It seem that this code is too old to run. Can anybody update recently?

Guys we are trying to run your code. The problem is this error below. I think your github is missing files. Can you please give us a workaround?:

FileNotFoundError Traceback (most recent call last) <ipython-input-18-b71253aaae34> in <module>() 156 txt_COVID='new_data/newtxt/train.txt', 157 txt_NonCOVID='/content/drive/My Drive/Covid_CT/Content2/CovidCTNew/Data-split/COVID/trainCT_NonCOVID.txt', --> 158 transform= train_transformer) 159 valset = CovidCTDataset(root_dir='new_data/4.4_image', 160 txt_COVID='new_data/newtxt/val.txt',

Hi,

Thank you for the for the effort of putting this dataset together.

Probably you have checked the metadata that is being collected for the covid x-ray dataset project. I would suggest on collecting a more specific and detailed metadata on each scan, as in the mentioned project, so that better insights can be obtained from the dataset.

Thanks!

Thanks for your work to help the people in need! Your site has been added to the Open-Source-COVID-19 page, which collects open source projects related to COVID-19, including maps, data, news, api, analysis, medical and supply information, etc. Please share to anyone who might need the information in the list, or will possibly contribute to some of those projects. You are also welcome to recommend more projects.

http://open-source-covid-19.weileizeng.com/

Cheers!

Hello, thank you very much for making this dataset available. I was looking at the non-covid images and I realized that even though they are classified as non-covid, a "crystallization" similar to covid is observed, do you know what other type of disease it is, if it is not covid?

Non-Covid:

Covid:

How to download dataset?

please delete this, I am sorry

Can you provide masks (ground glass opacity, consolidation) to evaluate a segmentation algorithm?

according to your readme file, your result was F1: 0.85，Accuracy: 0.86，AUC: 0.94，but mine was F1:0.74, Accuracy:0.77, AUC:0.88. Should I use the Self-Trans.pt as a pretrained model and fine tune on it?

It seems that the data split is with image-level, which means slices from the same patient could come to tran/test/val. This will bring some data-leakage problem. It is more reasonable to split in the patient level so that images from same patient cannot be assigned to different set(train/test/val)

Hello,

When I run the code 'DenseNet_predict.py', I think there is overfitting that the validation loss could not decrease properly without changing anything.

I got this trend when I run your code.

Could you please explain how to deal with this overfitting?

Thanks in advance.

For example, in the file Data-Splits/COVID/valCT_COVID.txt, there is a file called 2020.03.16.20036145-p19-128-4.png.

However, in the image processed folder, Image-Processed/COVID, the file is saved using a JPEG extension as 2020.03.16.20036145-p19-128-4.jpeg .

There are some other files like this too such as Images-processed/CT_COVID/2020.03.18.20038125-p15-54-2.png. I got over this by making sure all files had the same extension but please fix this.

Would Dicom files be part of future updates?

according to your code

self.classes = ['CT_COVID', 'CT_NonCOVID']
self.num_cls = len(self.classes)
self.img_list = []
for c in range(self.num_cls):
    cls_list = [[os.path.join(self.root_dir,self.classes[c],item), c] for item in read_txt(self.txt_path[c])]
    self.img_list += cls_list
self.transform = transform

the label of 'CT_COVID' is 0 and 'CT_NonCOVID' is 1, but when you calculate the metrics，like TP, your code is

TP = ((predlist == 1) & (targetlist == 1)).sum()

I think this is wrong since 'CT_COVID' is positive and its label is 0, we hope the model detect CT_COVID more accurate than CT_NonCOVID, so in this circumstance TP should be?

TP = ((predlist == 0) & (targetlist == 0)).sum()

Hi,

I am not sure how do you select the model to use for the test data. It seems to me in your code you did not use the information from the validation dataset to choose the model to use on the test data. I also did not observe any early stopping criteria being used. Is that true? If it is, how do you select the model and what is the purpose of the validation dataset?

Thanks,
Yangze

In your setting in 'DenseNet_predict.py', it seems you miss schedule.step() to change the learning rate. Could you please check this out?

It is a bit vague: does it include all non-COVID, i.e. Normal + Common Pneumonia, or only Common Pneumonia/only Normal?

Hi,
Thanks for the effort!
In the report, you only mentioned that the images of training set are collected from papers.
Can you guarantee that none of those are originally from SIRM?

Hi,

I have been spending some hours every other day to check for papers containing CT or Xray images of COVID-19 patients. Are you accepting contributions in any way? In the case of the covid-19 x-ray collection project, they keep a list of already collected papers, and one can inform of not-yet-collected papers in the issues section. Please advice us on how to contribute with papers containing CT images that you might have not yet included in your dataset.

FileNotFoundError: [Errno 2] No such file or directory: 'new_data/newtxt/train.txt'.

In Folder new_data and old_data/oldtxt/trainCT_NonCOVID.txt.
Is miss Train, val.txt and Test files.

NonCOVID-CT-MetaInfo.xlsx is not an excel ".xlsx" file but a csv.

Hi! Do you have any plan to upload NonCOVID-CT-Images.zip? Currently, only COVID-CT-Images.zip is available! Kindly advise. Thanks

The images with COVID come from various sources and most of them are annotated whereas the NON COVID images are all very homogeneous. If you train a classifier on theses images, you will build an "Image source classifier" , not a "COVID classifier"

Hello,

Thank you for contributing to research revolving around COVID-19. Your data set is very valuable.

I am not a physician, but upon consulting with collaborators it appears that the following issues should be addressed.

1. Soft tissue scans appear to be unpopular for this kind of application.
2. It appears that some cases of viral pneumonia are present in the control set. These will have similar (if not identical) manifestations to COVID-19.

Thank you again.