Apppublic

A-tavv/multimodal-image-registration

sourceHugging Faceupdated 10mo agoView on Hugging Face

0likes

App README

Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference

Multimodal CT/MRI Registration (SimpleITK)

This project implements a practical CT/MRI registration pipeline using SimpleITK/ITK:

—Preprocessing: resampling to common spacing, intensity normalization, histogram matching, and N4 bias correction.
—Registration: rigid then affine using Mattes Mutual Information with multi-resolution.
—Evaluation: Dice coefficient (if masks provided), overlay visualization, and deformation field inspection.

Setup

bash

python -m venv .venv
source .venv/Scripts/activate
pip install -r requirements.txt

Quick Start

bash

python mir.py \

Or create/edit config.json and simply run python mir.py; any CLI flags override the config values.

Bundled demo data lives under assets/sample_ct.nii.gz and assets/sample_mri.nii.gz; update config.json to point at those files for a ready-made example.

Features:

—Upload CT/MRI (NIfTI, MHA, PNG/JPG) and optional masks.
—Adjust target spacing + histogram matching.
—View overlay and deformation images instantly, download a ZIP of all outputs.
—Click “Load demo data” to preload the bundled synthetic volumes.

Using DICOM Directories

You can pass a directory instead of a single file for --fixed and/or --moving. The pipeline will automatically read the largest DICOM series. To target a specific series UID:

bash

python mir.py \
  --fixed "DICOM_CT_DIR" --fixed-series-uid "1.2.840...." \
  --moving "DICOM_MR_DIR" --moving-series-uid "1.2.840...." \
  --outdir outputs

If you omit --fixed-series-uid or --moving-series-uid, the series with the most slices is chosen.

Arguments --fixed-mask and --moving-mask are optional but recommended for Dice evaluation.

Outputs in outputs/:

—preprocessed/ fixed and moving images after normalization/matching.
—transforms/ initial + final transforms (.tfm).
—registered/ resampled moving aligned to fixed.
—eval/ metrics JSON, overlays PNG, and deformation field visualization.

Notes

—NIfTI (.nii/.nii.gz) or MHD/MHA recommended. DICOM series can be supported with minor changes.
—If histogram matching oversmooths CT air/bone, disable via --no-hist-match.
—DICOM reading uses SimpleITK's GDCM; ensure SimpleITK wheel includes GDCM (standard Windows wheels do).
—Pass --show to visualize the overlay interactively when the run finishes.
—Defaults in config.json let you run python mir.py with zero arguments; tweak paths or parameters there for quick demos.