lab-midas / med_segmentation Goto Github PK

Would it be possible to publish an example case such that we could see how to run this pipeline practically? For example just a folder containing a 3D data set, where when you run the pipeline the segmentation maps are produced? Would help a lot with user adoption.

Hi, thanks for sharing this work. I cloned this repository and read the paper "Fully automated and standardized segmentation of adipose tissue compartments via deep learning in 3D whole-body MRI of epidemiological cohort studies". Now I am wondering which of the models in ModelSet.py are used in the paper. Which one corresponds to the proposed DCNet and which one corresponds to the reference UNet?

I would like to use your trained model over my dataset. Would it be possible to use your DL model as a pre-trained model and do a transfer learning?

Hi Thomas,
Thank you for your code.
As I check the code has the model for landmark detection (body identification).
I want to apply this for cardiac (AV, Apex, Aorta root, mitral valve, left/right coronary)
but I don't know how to encode for above code? (is encoded as a mask or as X, Y, Z position?)

Thanks,
Tran

@all-contributors please add @marcfi for code, ideas, maintenance and tool

Hello, when I try to run this line: python3 main.py --preprocess -c config/config_default.yml I get the error

sage: MedSeg [-h] [-e EXP_NAME] [--preprocess PREPROCESS] [--train TRAIN] [--evaluate EVALUATE] [--predict PREDICT] [--restore RESTORE] [-c CONFIG_PATH] [--gpu GPU]
[--gpu_memory GPU_MEMORY] [--calculate_max_shape_only CALCULATE_MAX_SHAPE_ONLY] [--split_only SPLIT_ONLY] [--train_epoch TRAIN_EPOCH] [--filters FILTERS]
[--model_name MODEL_NAME] [--train_batch TRAIN_BATCH] [--dataset DATASET]
MedSeg: error: argument --preprocess: expected one argument

Any idea what is going on?

Hi Thomas,
Current setup.py is difference with requirements.txt
I think should change code as
from setuptools import setup, find_packages

`
import os

with open('./requirements.txt') as f:
requirements = f.read().splitlines()

setup(
name='med_segmentation',
version='1.0.0',
author='MIDAS and kSpace Astronauts',
author_email='[email protected]',
description='Medical Image Segmentation',
long_description=open(os.path.join(os.path.dirname(file), 'README.md')).read(),
package_dir={'med_seg': 'med_seg'},
packages=['med_seg'],
license='public',
keywords='None',
classifiers=[
'Natural Language :: English',
'Programming Language :: Python :: 3 :: Only'
],
install_requires==requirements,
)
`

Hi,
I try to run but get the error with missing function get_fixed_patches_index.
As I check pipeline.py and get_pad_and_patch.py is idendical.
Thanks.
Tran

Hello,
is there a pretrained model available? Most people do not have annotated datasets which they could use for training. They just want to generate tissue segmentations on their data. For this purpose a pretrained model which is ready to use would be great.

Hi Thomas,
I don't know why sorted function doesn't work for my dataset as it starts from negative

sorted(dicom_paths, key=lambda dicom_path: pydicom.dcmread(dicom_path).SliceLocation)

So I try by use numpy argsort as below.

    slicelocation = np.zeros(len(dicom_paths))
    # sort as slice location
    for f, i in zip(dicom_paths, range(len(dicom_paths))):
        ds = pydicom.read_file(f)
        if('slicelocation' in ds):	slicelocation[i] = ds.SliceLocation
        else:	slicelocation[i] = ds.ImagePositionPatient[2]
    order = np.argsort(slicelocation)
    dicom_paths= np.array(dicom_paths)[order]