Dataset Card for mri-sym2

Dataset Summary

SymBrain, an annotated dataset of brain MRI images designed to advance the field of brain symmetry detection and segmentation. Our dataset comprises a diverse collection of brain MRI T1w and T2w scans from the dHCP dataset. Each annotated to highlight the ideal straight mid-sagittal plane (MSP), demarcating the brain into two symmetrical hemispheres. The accurate extraction of the MSP has the potential to greatly enhance segmentation precision.

Researchers and practitioners can utilize this dataset to devise innovative methods for enhanced brain MRI image segmentation. SymBrain's rich and extensive content empowers the research community to address complex challenges in neuroimaging analysis, ultimately contributing to advancements in medical diagnostics and treatment planning.

Symmetry analysis plays an important role in medical image processing, particularly in the detection of diseases and malformations. SymBrain leverages the inherent bilateral symmetry observed in brain MRI images, making it an invaluable resource for the development and evaluation of automated algorithms aimed at detecting the symmetry axis within brain MRI data.

Dataset Structure

The dataset contains 1476 T1w images types and 1674 T2w images. The differences between the modalities lie in the intensity variations of the different brain areas. All the images are accessible in the 'train' part of the dataset.

Dataset Creation

Loading the data

The dataset contains a 'train' split of 1476 rows, containing the t1 type images, and a 'test' split of 1674 rows, with the t2 type images.

dataset = load_dataset("agucci/mri-sym2")
# first dataset example selection:
dataset['train'][0]

Attributes :

• image: PIL image, shape (290, 290)
• line: Straight line annotation coordinates on the image. ({'x':x1, 'y':y1}, {'x':x2, 'y':y2}). Where (x1,y1), (x2,y2) are the starting and end points of the line.
• radscore: Radiology score of the volume the image was extracted from. Please refer to dHCP doc for scores explanation.
• session: Session-ID of the original dataset, used for scan retrieval.

Source Data

dHCP dataset.
Three slices have been extracted from each of the 1050 3D volumes, creating 3150 images.

Annotations

The authors did Annotations manually with the V7lab tools.

Licensing Information

mit

Citation Information

When using the data please cite :

@misc{gucciardi2024symbrain,
      title={Symbrain: A large-scale dataset of MRI images for neonatal brain symmetry analysis}, 
      author={Arnaud Gucciardi and Safouane El Ghazouali and Francesca Venturini and Vida Groznik and Umberto Michelucci},
      year={2024},
      eprint={2401.11814},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}

and

dhcp dataset Data were provided by the developing Human Connectome Project, KCL-Imperial- Oxford Consortium funded by the European Research Council under the Eu- ropean Union Seventh Framework Programme (FP/2007-2013) / ERC Grant Agreement no. [319456]. We are grateful to the families who generously sup- ported this trial.