MedicalImageConverter 1.0.4

Reads in medical images and converts them into numpy arrays.

MedicalImageConverter

MedicalImageConverter is a Python package for working with medical image files. It lets the user read in images or ROIs (regions of interest), and converts them to 3D volumes. It also allows the user to input non-organized datasets (multiple images inside a single folder). Currently, this module only works for Dicom data with the hopes of expanding to other data formats in the future.

The module currently imports 9 different modalites and RTSTRUCT files. The accepted modalites are:

1. CT
2. MR
3. US
4. PT
5. MG
6. DX
7. NM
8. XA
9. CR

The CT and MR modalities have been tested extensively, along with their respective ROIs. The other 7 modalities have been tested but only on a few datasets a piece. For RTSTRUCTS, only those referencing CT and MR have been tested.

For sorting images or needing to output tags this module ideally works for any image orientation. However, if the user wants to output the 3D volumes of the image data or ROIs then the only orientation that currently works for is [1, 0, 0, 0, 1, 0] (this is the standard view of an axial image). It means the left side of the image is the right-side of the patient and the front of the image (top of image if you think in turns of 2D) is the anterior of the patient.

Disclaimer: All the files will be loaded into memory so be sure you have enough RAM available. Meaning don't select a folder path that contains 10s of different patient folders because you could possibly run out of RAM. Also, this module does not distinguish between patient IDs or patient names.

Installation

Using pip:

pip install MedicalImageConverter

Example

The user sets a path to the folder containing the dicom files. If the user already has each file path then they can continue to DicomReader, however if the user doesn't then use the file_parsar function. file_parsar will look through all the files inside the given file path and output a dictionary containing file paths for these types found:

1. Dicom (.dcm)
2. MetaHeader (.mhd)
3. Raw (.raw)
4. STL (.stl)

In the example below I used file_parsar, output the results into file_dictionary. Then the Dicom list was selected for the dictionary which was input into DicomReader class. Lastly, the data is then loaded in using load_dicom.

Note: The exclude_files allows the user not to get the file paths for any files inside the list, generally the user will use an empty list. The existing_image_info is required when the user is trying to load in an RTSTRUCT file only, some tags are needed for the image it references to create a 3D volume with the correct spacing.

import MedicalImageConverter as mic

path = r'/path/to/folder'

exclude_files = []
existing_image_info = None
file_dictionary = mic.file_parsar(path, exclude_files)
dicom_reader = mic.DicomReader(file_dictionary['Dicom'], existing_image_info)
dicom_reader.load_dicom()

Retrieve image information:

image_data = dicom_reader.get_image_data()  # Returns a list of 3D arrays containing each image
image_info = dicom_reader.get_image_info()  # Returns a pandas dataframe containing important tag information

Tags in image_info: FilePath, SOPInstanceUID, PatientID, PatientName, Modality, SeriesDescription, SeriesDate, SeriesTime, SeriesInstanceUID, SeriesNumber, AcquisitionNumber, SliceThickness, PixelSpacing, Rows, Columns, ImagePositionPatient, Slices, DefaultWindow

Retrieve ROI information:

roi_data = dicom_reader.get_roi_data()  # Returns a list of lists containing each 3D volume ROI per image
roi_info = dicom_reader.get_roi_info()  # Returns a pandas dataframe containing important tag information

Tags in roi_info: FilePath, RoiNames, PhysicalCoordinates, ArrayCoordinates

To save space the ROIs are stored in the smallest 3D volume binary possible, instead of the full image. So the roi_info tags are needed to reference where each ROI exist on an image.

Retrieve Sorted Files

ds_images = dicom_reader.get_ds_images()  # Returns a list of dicom files sorted into each image