tutorial.md

Welcome to this blog post on the simplicity of the Mhub API through Docker for elegant and reproducible medical imaging processing.

Today, we're diving into the C4KC-KITS dataset—a curated collection of CT scans centered on kidney tumors. Each study comes packed with DICOM-formatted images, detailed metadata, segmentation masks, and clinical annotations. This rich repository not only accelerates computer-aided diagnosis research but also fuels the development and benchmarking of cutting-edge machine learning models in cancer imaging.

The magic begins with downloading your dataset using the Imaging Data Commons API. Simply execute:

idc download c4kc_kits

This command retrieves all the necessary DICOM files. If you also need metadata, remember you can fetch it from TCIA: https://www.cancerimagingarchive.net/collection/c4kc-kits

Once downloaded, your data will look similar to this:

suraj@R2-Z9:/mnt/data1/datasets/RadiomicsHub/C4KC_KiTS/c4kc_kits$ ls c4kc_kits
KiTS-00000  KiTS-00018  KiTS-00036  KiTS-00054  KiTS-00072  KiTS-00090  KiTS-00108  KiTS-00126  KiTS-00144  KiTS-00162  KiTS-00180  KiTS-00198
KiTS-00001  KiTS-00019  KiTS-00037  KiTS-00055  KiTS-00073  KiTS-00091  KiTS-00109  KiTS-00127  KiTS-00145  KiTS-00163  KiTS-00181  KiTS-00199
...

Transitioning from DICOM to DL-friendly formats like NIfTI or NRRD is traditionally a challenging task, but this is where the elegance of Mhub truly shines. The Mhub API, exposed via Docker, provides a simplified, robust, and elegant solution to convert, organize, and process your imaging data with minimal fuss.

Start by pulling the latest Mhub base container:

docker pull mhubai/base:latest

Next, set up your processing configuration with a YAML file that directs the Mhub API precisely how to handle your data:

general:
  data_base_dir: /app/data
  version: 1.0
  description: Sort DICOM files into a structured directory

execute:
  - DicomImporter
  - DsegExtractor
  - NiftiConverter
  - DataOrganizer

modules:
  DicomImporter:
    source_dir: input_data
    import_dir: sorted_data
    sort_data: true
    merge: true
    meta: 
      mod: '%Modality'
      desc: '%SeriesDescription'

  DataOrganizer:
    targets:
    - nifti:mod=ct-->[i:sid]/ct.nii.gz  # Organize CT images by series ID
    - nifti:mod=seg-->[i:sid]/[d:roi]_seg.nii.gz  # Prefix segmentation files with ROI information

The Mhub API comes to life with a single Docker command, making use of volume mappings to direct your input and output data paths. Execute:

docker run -v /mnt/data1/datasets/RadiomicsHub/C4KC_KiTS/c4kc_kits:/app/data/input_data \
-v /mnt/data1/datasets/RadiomicsHub/C4KC_KiTS/c4kc_kits_processed:/app/data/output_data \
-v /home/suraj/Repositories/config:/app/utility/config -it mhubai/base:latest --utility

This command leverages the full power of the Mhub API to process your DICOM data effortlessly. After successful execution, you’ll find your output directory structured neatly:

....
....
├── 1.3.6.1.4.1.14519.5.2.1.6919.4624.927655669179864696894393596289
│   ├── ct.nii.gz
│   ├── Kidney_seg.nii.gz
│   └── Renal Tumor_seg.nii.gz
├── 1.3.6.1.4.1.14519.5.2.1.6919.4624.933941010667753097000370321589
│   ├── ct.nii.gz
│   ├── Kidney_seg.nii.gz
│   └── Renal Tumor_seg.nii.gz
...

Thanks to the elegant design of the Mhub API, powered by Docker, converting DICOM images and extracting segmentation masks is now a streamlined, hassle-free experience. With a few commands and a simple config file, you’re all set to focus on your research rather than the nitty-gritty of data conversion. Embrace the simplicity of Mhub and let Docker handle the heavy lifting for you!