KinderSeg: FastSurfer Database for Age-Specific Brain Volumes in Healthy Children

Code for the paper:

 Ibrahim Zughayyar, Martin Bauer, Christopher Güttler, Ana Luísa Marcelino, Fabienne Kühne, Claudia Buss, Christine Heim, Annette Aigner, 
 Anna Tietze, and Andrea Dell'Orco 
(2024, September 9).
A FastSurfer Database for Age-Specific Brain Volumes in Healthy Children:
A Tool for Quantifying Localized and Global Brain Volume Alterations in Pediatric Patients.

https://doi.org/10.31219/osf.io/dw7p4

https://osf.io/kj7hy/

Run with Docker

git clone https://github.com/ibrazug/kinderseg.git
cd kinderseg/Docker
cp </path/to/your/license.txt> .
docker build -t kinderseg .
docker run --gpus all --rm \
    -v <nifti_data>:/data \
    -v <output_dir>:/output \
    kinderseg \
    --age <age> \
    --threads <threads, optional, default=4>

IMPORTANT NOTE: --threads is used to define the number of threads in total. not per hemisphere.

nifti_data can be either a directory containing NIfTI files or a single NIfTI file. The output will be saved in the output directory in a subdirectory named after the input NIfTI file, without the extension. Example:

docker run --gpus all --rm \
    -v $PWD/sub-001/ses-01/anat/sub-001_ses-01_T1w.nii.gz:/data/sub-001_ses-01_T1w.nii.gz \
    -v ./output:/output \
    kinderseg --age 12

Run with Apptainer/Singularity

git clone https://github.com/ibrazug/kinderseg.git
cd kinderseg/Docker
cp </path/to/your/license.txt> .
docker build -t kinderseg .
apptainer build kinderseg.sif docker-daemon://kinderseg:latest 
apptainer run --nv kinderseg.sif \
    -B <nifti_data>:/data \
    -B <output_dir>:/output \
    --age <age> \
    --threads <threads>

Requirements

• At least 5 GB of RAM
• GPU (optional, but recommended)
• a valid FreeSurfer license: you can get one for free by clicking here.
• we also provided Docker.amd version for AMD chips.

Dataset Statistics

Dataset	N° in total	Ex (QC)	N° after QC	Ex (SC)	N° after SC
HBN	170	21	149	10	139
LOC	125	32	93	1	92
Kids2Health	211	5	206	1	205
Total	506	58	448	12	436

MRI Segmentation and Volumetric Analysis of the dataset subjects

MRI Post-processing

• DICOM data converted to NIfTI format using dcm2niix and Python version 3.9 was used for data manipulation

Segmentation Tools

1. FreeSurfer v7.4: Ran on a Slurm cluster with a 16-core Intel Xeon E5-2650 CPU, Utilized GNU Parallel for parallel processing

2. FastSurfer v2.3.3: Utilized FastSurferCNN pipeline on a workstation with an AMD 3970X CPU and NVIDIA GeForce RTX 3090 GPU

Data Analysis

• R version 4.1.2 used for analysis
• Libraries: ggplot2, tidyverse, viridis (R)

KinderSeg Module Files

The kinderseg directory contains the following key files and scripts:

1. metrics.py:

   • Implements metrics and evaluation methods for comparing segmented brain volumes.
   • Includes the DiceNiiRois class for calculating the Dice similarity coefficient between ROIs in NIfTI files.
   • Outputs a Pandas DataFrame with the Dice similarity between all ROIs.

2. sums.py:

   • Contains utility classes for summarizing region of interest (ROI) data and volume tables.
   • Classes:
     • SumNiiRois: Summarizes ROIs as defined in the sumroi_mapping.csv file and generates output NIfTI files.
     • SumVolTable: Summarizes volume tables (from FastSurfer and FreeSurfer) based on mapped ROIs, creating CSV output.
   • Uses parallel processing to speed up calculations and ROI summarization.

3. sumroi_mapping.csv:

   • A CSV file that contains mappings between raw ROIs and summarized ROIs.
   • Supports mappings from VINN and FreeSurfer 7 (FS7) formats.
   • Used by sums.py to transform and summarize volume data efficiently.