Breast MRI Preprocessing Pipeline

This repository contains a preprocessing pipeline for breast MRI data, designed to prepare medical images for AI model analysis. The pipeline handles DICOM to NIfTI conversion and image coregistration for multiple MRI sequences.

Overview

The pipeline processes breast MRI data with multiple sequences (T1WI, T2WI, DWI, ADC) and includes the following steps:

1. DICOM to NIfTI conversion
2. Normalization of image intensities
3. Image coregistration (optional)
4. Automatic organization of processed files

Input Data Structure

The expected input data structure is as follows:

└─Input_Directory
    ├─PatientID_1
    │  ├─ADC_map
    │  ├─DWI
    │  ├─FS_T1WI
    │  ├─NonFS_T1WI
    │  ├─reference_CE_MRI
    │  └─T2WI
    ├─PatientID_2
    │  ├─ADC_map
    │  ├─DWI
    │  ├─FS_T1WI
    │  └─...
    └─...

Each patient directory contains multiple MRI sequences in DICOM format:

• FS_T1WI: Fat-suppressed T1-weighted images
• NonFS_T1WI: Non-fat-suppressed T1-weighted images
• T2WI: T2-weighted images
• DWI: Diffusion-weighted images (includes multiple b-values: b0, b800, b1200)
• ADC_map: Apparent Diffusion Coefficient maps
• reference_CE_MRI: Contrast-enhanced reference images

Output Structure

The processed data is organized as follows:

└─Output_Directory
    ├─input
    │   ├─b800
    │   │   └─PatientID
    │   │       ├─PatientID_T1FS_wm.nii.gz
    │   │       ├─PatientID_T2FS_wm.nii.gz
    │   │       ├─PatientID_ADC_wm.nii.gz
    │   │       ├─PatientID_B0_wm.nii.gz
    │   │       └─PatientID_DWI_wm.nii.gz
    │   └─b1200
    │       └─PatientID
    │           └─...
    └─output
        ├─b800
        │   └─PatientID
        │       └─reference_CE_MRI.nii.gz
        └─b1200
            └─PatientID
                └─reference_CE_MRI.nii.gz

Features

• Automatic DICOM to NIfTI conversion
• Handles multiple MRI sequences
• Special processing for DWI sequences with different b-values
• Image intensity normalization
• Optional image coregistration (supports rigid and affine transformations)
• Organized output structure for AI model input
• Progress tracking with tqdm
• Comprehensive error handling

Requirements

numpy
antspy
SimpleITK
pydicom
tqdm

Installation

git clone https://github.com/yourusername/breast-mri-preprocessing.git
cd breast-mri-preprocessing
pip install -r requirements.txt

Usage

Basic usage:

from mri_preprocessing_ver_01 import DCM2NIIConverter, Coregistration

# Convert DICOM to NIfTI
input_dir = '/path/to/dicom/data'
converter = DCM2NIIConverter(input_dir)
converter.convert_all()

# Perform coregistration
output_dir = '/path/to/output'
coregistration = Coregistration(output_dir)

# Without registration (default)
coregistration.process_all()

# With registration (optional)
coregistration.process_all(registration_type='rigid')  # or 'affine' or 'both'

Registration Options

The pipeline supports different registration types:

• No registration (default): Only performs interpolation
• Rigid registration: Corrects for rotation and translation
• Affine registration: Corrects for rotation, translation, scaling, and shearing
• Both: Applies both rigid and affine transformations sequentially

Note

• The pipeline uses FS T1WI as the reference image for coregistration
• All images are normalized to the range [0,1]
• DWI sequences are automatically sorted by b-values
• Progress bars show processing status for each step