Conversion of ImageObject to NIfTI

[MLK97]

After several hours of trying and looking through all kind of examples and tutorials I am close to giving up onthis simple problem.

I want to make a demo app in Monai that does nothing more than convert a dicom series to a nifti image.
For this, I use a workflow that looks like this

        # DICOM Data Loader Operators
        loader_op = DICOMDataLoaderOperator(self, CountCondition(self, 1), input_folder=input_data_path, name="loader_op")
        series_selector_op = DICOMSeriesSelectorOperator(self, rules=SERIES_RULE, name="series_selector_op")

        # DICOM to NIfTI
        dicom_to_volume_op = DICOMSeriesToVolumeOperator(self, name="dicom_to_volume_op")
        volume_to_nifti_op = VolumeToNiftiOperator(self, name="volume_to_nifti_op")


        # Operator pipeline
        self.add_flow(loader_op, series_selector_op, {("dicom_study_list", "dicom_study_list")})
        self.add_flow(series_selector_op, dicom_to_volume_op, {("study_selected_series_list", "study_selected_series_list")})
        self.add_flow(dicom_to_volume_op, volume_to_nifti_op,{("image", "image_volume")})

where DICOMDataLoaderOperator, DICOMSeriesSelectorOperator and DICOMSeriesToVolume are just the pre-defined operators found in monai-deploy.

The VolumeToNiftiOperator looks like this:

import logging

from monai.deploy.core import ConditionType, Fragment, Operator, OperatorSpec
from monai.deploy.core.domain.dicom_series_selection import StudySelectedSeries
from monai.deploy.core.domain.image import Image
from monai.transforms import SaveImage


class VolumeToNiftiOperator(Operator):
    """This operator converts an instance of DICOMSeries into a NIfTI image.

    The loaded Image Object can be used for further processing via other operators.

    Named Input:
        input_volume: Image Object from Monai Deploy.
    Named Output:
        image_nifti_path: filepath to nifti image.
    """

    def __init__(self, fragment: Fragment, *args, **kwargs):
        """Create an instance for a containing application object.

        Args:
            fragment (Fragment): An instance of the Application class which is derived from Fragment.
        """

        self.input_volume = "image_volume"
        self.output_nifti = "image_nifti_path"
        # Need to call the base class constructor last
        super().__init__(fragment, *args, **kwargs)

    def setup(self, spec: OperatorSpec):
        spec.input(self.input_volume)
        spec.output(self.output_nifti).condition(ConditionType.NONE)
        
    def compute(self, op_input, op_output, context):
        """Performs computation for this operator and handles I/O."""
        logging.debug('Starting Volume to Nifti Operator')
        volume_image = op_input.receive(self.input_volume)
        
        saver = SaveImage(
            output_dir='./', 
            output_postfix='_0000', 
            output_ext='.nii.gz',
            writer="NibabelWriter" ,
            resample=False)
        saver(volume_image)

        placeholder = 'placeholder'
        op_output.emit(placeholder, self.output_nifti)
        logging.debug('Ending Volume to Nifti Operator')

The code executes when I run the method saver with just the image information, that is, saver(volume_image.asnumpy()) but I would like to run this with the metadata contained in volume_image.metadata() when I try to run this as written in the code above I get an error:

Exception occurred for operator: 'volume_to_nifti_op'
Traceback (most recent call last):
  File "/home/ai/monai_nnUNetv2_ProstateMRL_map/operators/volume_to_nifti_operator.py", line 63, in compute
    saver(volume_image)
  File "/home/ai/miniconda3/envs/ProstateMRL_MONAI/lib/python3.10/site-packages/monai/transforms/io/array.py", line 500, in __call__
    raise RuntimeError(
RuntimeError: SaveImage cannot find a suitable writer for test/0/0__0000.nii.gz.
    Please install the writer libraries, see also the installation instructions:
    https://docs.monai.io/en/latest/installation.html#installing-the-recommended-dependencies.
   The current registered writers for .nii.gz: (<class 'monai.data.image_writer.NibabelWriter'>,).
Traceback (most recent call last):
  File "/home/ai/miniconda3/envs/ProstateMRL_MONAI/lib/python3.10/site-packages/monai/transforms/io/array.py", line 484, in __call__
    writer_obj.set_data_array(data_array=img, **self.data_kwargs)
  File "/home/ai/miniconda3/envs/ProstateMRL_MONAI/lib/python3.10/site-packages/monai/data/image_writer.py", line 580, in set_data_array
    self.data_obj = self.convert_to_channel_last(
  File "/home/ai/miniconda3/envs/ProstateMRL_MONAI/lib/python3.10/site-packages/monai/data/image_writer.py", line 316, in convert_to_channel_last
    data = moveaxis(data, _chns, tuple(range(-len(_chns), 0)))
  File "/home/ai/miniconda3/envs/ProstateMRL_MONAI/lib/python3.10/site-packages/monai/transforms/utils_pytorch_numpy_unification.py", line 70, in moveaxis
    return np.moveaxis(x, src, dst)
  File "<__array_function__ internals>", line 200, in moveaxis
  File "/home/ai/.local/lib/python3.10/site-packages/numpy/core/numeric.py", line 1467, in moveaxis
    source = normalize_axis_tuple(source, a.ndim, 'source')
  File "/home/ai/.local/lib/python3.10/site-packages/numpy/core/numeric.py", line 1398, in normalize_axis_tuple
    axis = tuple([normalize_axis_index(ax, ndim, argname) for ax in axis])
  File "/home/ai/.local/lib/python3.10/site-packages/numpy/core/numeric.py", line 1398, in <listcomp>
    axis = tuple([normalize_axis_index(ax, ndim, argname) for ax in axis])
numpy.AxisError: source: axis 0 is out of bounds for array of dimension 0

I am rather sure that my image is healthy and in good condition as the operators handling the Dicom image before it langs in the VolumeToNiftiOperator work fine.
It is extremely hard to find a correct way of doing this so the solution might be obvious but as a novice in the framework it is very hard to really understand how these methods are supposed to be used, especially since it seems like solutions for versions prior to 0.6.0 do not apply and work in this version anymore.

[KumoLiu]

Hi @MLK97, I think LoadImage can load DICOM directly.
https://github.com/Project-MONAI/MONAI/blob/6c9e49ee045c345ecd8d3518cb0d503ed02c61b7/monai/transforms/io/array.py#L101
https://github.com/Project-MONAI/MONAI/blob/6c9e49ee045c345ecd8d3518cb0d503ed02c61b7/monai/data/image_reader.py#L376

Thanks.

[MLK97]

Thanks for the fast reply! It does indeed but it does not take care of the entire DicomSort and SeriesSelection. That is where the operators are very useful.
Also in general I am rather interested to find out what the problem is when I want to save a MONAI Image Object.

[KumoLiu]

Did you ensure the data is channel-first?
https://github.com/Project-MONAI/MONAI/blob/6c9e49ee045c345ecd8d3518cb0d503ed02c61b7/monai/transforms/io/array.py#L469

[MLK97]

This could be it. Thanks. I will try it as soon as I have access to the development environment where the project is!

[MLK97]

Thank you that worked. But there is also a difference between the code that pip installed and the documentation of the 'latest' version.
The installed monai version is 1.3.0 and looking at the __call__ method of that version there is no filename parameter
https://github.com/Project-MONAI/MONAI/blob/main/monai/transforms/io/array.py

However, the latest version of the documentation still has this parameter. Is this an issue with updating the documentation?
https://docs.monai.io/en/latest/transforms.html

[KumoLiu]

Hi @MLK97, the filename support is introduced after v1.3.
Project-MONAI/MONAI#7318
Thanks.