Running testing.py

[build2create]

Will be this be correct way to the testing.py

if __name__ == '__main__':
    run_test_case(0,"./data_test")

where ./data_test is the name to the out_dir where the prediction.nii.gz should I also call predict_from_data_file_and_write_image from run_test_case() to actually write the image?

[ellisdg]

@build2create sorry, I hadn't tested that code out yet. With the latest commit that I just committed, you should be able to run the code as you have it in your comment.

[build2create]

Alright!

[build2create]

With the pre-trained model given here I generated training_ids.pklandtesting_ids.pkl.Bascially just to get testing ids required to test the model. I defined main class as :

if __name__ == '__main__':
    run_test_case(1,"./data_test")

The problem is it is not producing any data_dir .Neither it is appending to it if I create one. And this:

if __name__ == '__main__':
    model = load_model(config["model_file"])
    predict_from_data_file_and_write_image(model,config["hdf5_file"],0,result.nii.gz)

And this

if __name__ == '__main__':
    model = load_model(config["model_file"])
    predict_from_data_file_and_write_image(model,config["hdf5_file"],0,data_dir)#data_dir  is directory

All of them fail to produce any segmented output. Plus where is get_prediction_labels getting called? @ellisdg any idea on this one? Apologies for the constant trouble.

[ellisdg]

Maybe try giving a full path name instead of "./data_test"? Does that work?

[build2create]

It worked but gave error ValueError: "concat" mode can only merge layers with matching output shapes except for the concat axis. Layer shapes: [(None, 1024, 18, 18, 9), (None, 256, 18, 18, 18)] then I updated the configuration in keras.json file as

{
"image_dim_ordering": "th",
"epsilon": 1e-07,
"floatx": "float32",
"backend": "tensorflow"
}

Note: earlier"backend": "theano" this was the in the keras.json file so it was giving error.(refer this) it ran finally but gave error: ValueError: Invalid objective: dice_coef_loss

then I added these: from model import dice_coef_loss from model import dice_coef and model = load_model(config["model_file"],custom_objects={'dice_coef_loss': dice_coef_loss,'dice_coef':dice_coef})

Now I get this error:

ValueError: Error when checking : expected input_1 to have 5 dimensions, but got array with shape (3, 144, 144, 144) To which I added 'T2' to config["training_modalities"] Now to the call in main: run_test_case(1,'/home/adminsters/Desktop/3DUnetCNN-master/data_dir'); it gives index error:

  File "testing.py", line 67, in <module>
    run_test_case(1,'/home/adminsters/Desktop/3DUnetCNN-master/data_dir');
  File "testing.py", line 53, in run_test_case
    image = nib.Nifti1Image(test_data[i], affine)
IndexError: index 3 is out of bounds for axis 0 with size 3

My testing_ids.pkl is: [141, 20, 15, 57, 186, 167, 60, 80, 23, 54, 143, 158, 132, 83, 61, 81, 168, 85, 157, 90, 147, 86, 127, 199, 74, 136, 24, 182, 146, 152, 170, 44, 30, 156, 172, 154, 1, 25, 59, 45]

I guess it is because I changed the config["training_modalities"] @ellisdg how to fix this Index error?

[ellisdg]

There are no error messages? Are you sure it is running? Have you tried a debugger?

On Apr 20, 2017 11:47 PM, "build2create" notifications@github.com wrote:

Nope, it did not work

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/ellisdg/3DUnetCNN/issues/7#issuecomment-296062346, or mute the thread https://github.com/notifications/unsubscribe-auth/AIKZk4TpqlJ5VukIkX3dhc5wNrdF5yI_ks5ryDT1gaJpZM4M6908 .

[build2create]

I am getting index error. Please see updated comment

[build2create]

In case you are not able to see I am reposting it here: Initially I was getting this: ValueError: "concat" mode can only merge layers with matching output shapes except for the concat axis. Layer shapes: [(None, 1024, 18, 18, 9), (None, 256, 18, 18, 18)] then I updated the configuration in keras.json file as

{
"image_dim_ordering": "th",
"epsilon": 1e-07,
"floatx": "float32",
"backend": "tensorflow"
}

Note: earlier"backend": "theano" this was the in the keras.json file so it was giving error.(refer this) it ran finally but gave error: ValueError: Invalid objective: dice_coef_loss then I added these:

from model import dice_coef_loss
from model import dice_coef

and model = load_model(config["model_file"],custom_objects={'dice_coef_loss': dice_coef_loss,'dice_coef':dice_coef}) Now I get this error:

ValueError: Error when checking : expected input_1 to have 5 dimensions, but got array with shape (3, 144, 144, 144) So in config.py I added 'T2' to config["training_modalities"] Now to the call in main: run_test_case(1,'/home/adminsters/Desktop/3DUnetCNN-master/data_dir'); it gives index error:

  File "testing.py", line 67, in <module>
    run_test_case(1,'/home/adminsters/Desktop/3DUnetCNN-master/data_dir');
  File "testing.py", line 53, in run_test_case
    image = nib.Nifti1Image(test_data[i], affine)
IndexError: index 3 is out of bounds for axis 0 with size 3

My testing_ids.pkl is: [141, 20, 15, 57, 186, 167, 60, 80, 23, 54, 143, 158, 132, 83, 61, 81, 168, 85, 157, 90, 147, 86, 127, 199, 74, 136, 24, 182, 146, 152, 170, 44, 30, 156, 172, 154, 1, 25, 59, 45]

I guess it is because I changed theconfig["training_modalities"] @ellisdg how to fix this Index error?

[ellisdg]

It looks like you may have modified line 53, or you may not have the latest version of the file.

[build2create]

Ok my bad, ran the updated code this time. I got his output : for a test image id 20. How should I infer segmentation labels from this, which one is necrotic,edemic and so on? Like the one in this When I choose to open segmented image I get this: Which patch is tumorous how should I interpret?

[ellisdg]

The image is a scalar image and not a label map. You need to view it in grayscale as a scalar image.

On Apr 22, 2017 12:24 AM, "build2create" notifications@github.com wrote:

Ok my bad, ran the updated code this time. I got his output : [image: image] https://cloud.githubusercontent.com/assets/25721143/25301595/219251ba-2749-11e7-87a8-0dd95880fda5.png for a test image id 20. How should I infer segmentation labels from this, which one is necrotic,edemic and so on? Like the one in this https://github.com/naldeborgh7575/brain_segmentation When I choose to open segmented image I get this: [image: image] https://cloud.githubusercontent.com/assets/25721143/25301629/d02a263a-2749-11e7-86a1-ddaf4b51fbb8.png Which patch is tumorous how should I interpret?

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/ellisdg/3DUnetCNN/issues/7#issuecomment-296349055, or mute the thread https://github.com/notifications/unsubscribe-auth/AIKZkyAJ_X9es17FxQyJpYfJtyBvOcdYks5ryY76gaJpZM4M6908 .

[build2create]

Can you elaborate? The first screenshot shows grayscale image. Is that correct?How do I read the first image?Is it like darker parts tumorous? Can you cite/give an example?

[build2create]

If I have understood it right this is not for tumor detection but just for volumetric segmentation right?

[build2create]

For FLAIR modality I am sort of getting evidently slightly more greyish patch, Is that the tumor?

[ellisdg]

T1: T1c: FLAIR: Tumor prediction. Ranges from 0 to 1: Manual segmentation. The separate tumor regions are not used by the classifier:

[build2create]

Thanks. By the way which tool are you using for viewing? Is it ITK sanp?

[ellisdg]

3D Slicer. I know a lot of people use ITK Snap, but I've never taken the time to learn how to use it. 3D Slicer has a lot of features, so there might be a little bit of a learning curve. They also have plenty of helpful tutorials too.

[build2create]

Thanks a lot!

[build2create]

By volumetric segmentation we are detecting the presence or absence of the tumor but not the different segments within the tumor, like necrotic, core or enhancing. Doing this would be a multi-class classification problem right?

[build2create]

Also why are we not able to segment the tumor into sub-parts.Is it because of data set?I viewed ground truth of BRATS dataset in mha format. They do provide all the segmentation labels.

[ellisdg]

You should be able to train a classifier to segment different tumor regions with Keras. However, I currently have no need for multi-class labels, so I have not implemented it yet. I will probably add this eventually, but that could be months from now.

[mingrui]

for anyone coming from the future, testing.py was renamed to predict.py:

https://github.com/ellisdg/3DUnetCNN/commit/b4012bb0b74dc7c48aa468fc0d277f5f0e374720#diff-22508acf6c6356eb94fbcda75eaf09f2