Summary

Adds a beginner-friendly introductory tutorial for MRI reconstruction using the fastMRI knee single-coil dataset and MONAI's reconstruction transforms.

Changes

• New notebook: reconstruction/MRI_reconstruction/tutorials/01_kspace_basics_fastmri_knee.ipynb
  • Part 1: What is k-space (loading real data, inverse FFT)
  • Part 2: Fourier transform connection (low vs. high frequencies)
  • Part 3: Undersampling and aliasing artifacts (1x, 2x, 4x, 8x)
  • Part 4: Random vs. equispaced masks using RandomKspaceMaskd / EquispacedKspaceMaskd
  • Part 5: Full MONAI preprocessing pipeline (FastMRIReader → CenterSpatialCropd → ReferenceBasedNormalizeIntensityd)
  • Part 6: Zero-filled reconstruction → deep learning connection
• New README: reconstruction/MRI_reconstruction/tutorials/README.md
• Updated README.md: Added missing Reconstruction section listing this tutorial plus existing unet_demo and varnet_demo
• Updated runner.sh: Added notebook to doesnt_contain_max_epochs (no training loop)

Dataset

• Uses fastMRI knee single-coil validation set (non-commercial license)
• Only one .h5 file (~300 MB) required — no need for the full ~1.5 TB brain multi-coil download
• Added to skip_run_papermill via existing .*MRI_reconstruction.* pattern

Design decisions

• Bridges the gap between newcomers and production tutorials (unet_demo, varnet_demo)
• Uses CenterSpatialCropd instead of ReferenceBasedSpatialCropd to handle the k-space (640×368) vs ground truth (320×320) dimension mismatch
• No training loop — purely educational, runs without GPU

Test plan

• Notebook runs end-to-end on Google Colab with real fastMRI knee data (file1000000.h5)
• All 5 matplotlib visualizations render correctly
• Pipeline output shapes: kspace_masked_ifft and reconstruction_esc both (1, 320, 320)
• PEP 8 passes via ./runner.sh -t reconstruction/MRI_reconstruction/tutorials/01_kspace_basics_fastmri_knee.ipynb --no-run
• Copyright header present with correct formatting