MISGRA: Meningioma Imaging with Spatially Grounded Report Annotations

Overview

MISGRA is a multi-modal, longitudinal dataset of brain MRI scans, de-identified radiology reports, and expert spatial annotations from 103 patients diagnosed with meningioma at King's College Hospital (KCH), London, UK.

The dataset provides spatial grounding of radiology report findings on 3D MRI volumes, linking clinical text descriptions to their anatomical locations. Annotations were created by five senior neurosurgery residents using a custom 3D Slicer extension.

This repository contains the code used to curate the dataset and reproduce the validation analyses reported in the accompanying data descriptor paper. The dataset itself is available on Synapse.

Dataset at a Glance

Characteristic	Value
Patients	103
Scan sessions	433
Total MRI volumes	1,199
Radiology reports	433
Expert annotations	2,772
MRI sequences	ceT1, T1, T2, FLAIR
Sessions per patient (mean +/- SD)	4.20 +/- 3.08
Patients with multiple sessions	82 (79.6%)
Age at first scan - median (IQR)	57 (44-70) years
Sex (female / male)	73 (70.9%) / 30 (29.1%)

For full dataset documentation (structure, file formats, metadata schema), see the README included with the dataset on Synapse.

Code

Requirements

• Python >= 3.8
• numpy, pandas, openpyxl

pip install -r requirements.txt

Scripts

Script	Description
nifti_mri_classifier.py	Classifies MRI sequences from DICOM-derived folder names using keyword matching
build_misgra_dataset.py	Selects the best scan per modality per session and copies to the release directory with provenance metadata
misgra_dataset_summary.py	Generates summary statistics tables for the released dataset
intra_annotator_agreement.py	Computes intra-annotator agreement (Euclidean distance, Dice, ICC) from paired re-annotation sessions
inter_annotator_agreement.py	Computes inter-annotator agreement across five annotators using pairwise comparisons and multi-rater ICC

Usage

# Classify MRI sequences from folder names
python nifti_mri_classifier.py /path/to/nifti/directory -o classification_results.csv

# Build the release dataset (select best scans per session)
python build_misgra_dataset.py \
    --given-dir /path/to/annotator_folders \
    --xnat-dir /path/to/source_nifti \
    --validation-csv /path/to/validation_results.csv \
    --output-dir /path/to/output

# Generate dataset summary statistics
python misgra_dataset_summary.py --misgra-dir /path/to/MISGRA --output-dir ./tables

# Compute intra-annotator agreement
python intra_annotator_agreement.py \
    --spreadsheet /path/to/Intra_annotator.xlsx \
    --annotations-root /path/to/annotations \
    --output-csv intra_results.csv

# Compute inter-annotator agreement
python inter_annotator_agreement.py \
    --spreadsheet /path/to/Inter_annotator.xlsx \
    --annotations-root /path/to/annotations \
    --output-csv inter_results.csv

Data Availability

The MISGRA dataset is available on Synapse at syn76420718. Download will be allowed soon.

Ethics

This study was approved by the NHS Research Ethics Committee (REC reference: 22/NS/0160; IRAS ID: 320754).

Citation

If you use this dataset, please cite:

Garcia-Foncillas Macias, L., Benjamin, P., Elshalakany, A., Wroewright, O., Kalaitzoglou, D., Anagnostou, E., Awan, M., Kalyal, N., Vercauteren, T., & Shapey, J. (2025). MISGRA: Meningioma Imaging with Spatially Grounded Report Annotations [Dataset]. Synapse. https://www.synapse.org/Synapse:syn76420718

Contact

• Lorena Garcia-Foncillas Macias (lorena.garcia-foncillas_macias@kcl.ac.uk) — King's College London, UK
• Jonathan Shapey (jonathan.shapey@kcl.ac.uk) — King's College London, UK
• Tom Vercauteren (tom.vercauteren@kcl.ac.uk) — King's College London, UK

Licence

• Dataset: CC BY-NC 4.0
• Code: Apache 2.0