Multiparametric MRI, manual segmentations, radiomics and molecular profiles of diffuse gliomas

Abstract

This dataset contains de-identified preoperative and postoperative multiparametric brain MRI examinations from 192 adult patients with diffuse gliomas (WHO CNS5 grades 2–4), including 156 preoperative and 161 postoperative examinations. The release includes multicompartment tumour-region segmentations, clinical and molecular variables, radiomics feature tables, time-to-progression data, and overall-survival data.

Key points

• The cohort originates from Central Europe and broadens the geographic representation of publicly available glioma datasets.

• Both de-identified raw MRI and preprocessed, co-registered MRI are provided, together with manual segmentations and radiomics features.

• The resource supports radiogenomic research, segmentation benchmarking, and studies of imaging-based outcome prediction.

• Users should account for the observational, single-cohort nature of the dataset, imaging heterogeneity, and incomplete molecular testing in some cases.

Technical information

The release is organized into six components.

1. Clinical and molecular data

1_0_clinical_molecular_data.csv contains patient-level clinical and molecular variables. The id column links the tabular data to imaging, segmentation masks, and radiomics data.

Variable definitions and permitted values are provided in:

1_1_clinical_molecular_metadata_dictionary.csv

The molecular variables include recorded results for:

• IDH1/2

• ATRX

• TP53

• TERT promoter

• 1p/19q co-deletion

• CDKN2A/B

• EGFR

• chromosomal alterations, including chromosome 7 gain and chromosome 10 loss

Molecular testing was not available for every marker in every patient. Marker-level reasons for missingness were not retained for this retrospective release.

Integrated diagnoses follow the released pathology and molecular-support fields. The following columns document whether the available molecular evidence completely supports, partially supports, or does not support the reported WHO CNS5 diagnostic suffix:

• diagnosis_family_for_support_rule

• molecular_support_flag

• who_cns5_suffix_as_reported

• molecular_support_rule

The release identifies:

• 18 cases as NOS_as_reported

• 1 case as NEC_as_reported

• 8 cases with partial_molecular_support

Users should use these fields when selecting cases for diagnosis-specific analyses and should not infer an unrecorded molecular result from a missing value.

2. MRI imaging

The imaging archives contain the following sequences where available:

• T2-weighted (T2w)

• FLAIR (T2f)

• non-contrast T1-weighted (T1n)

• delayed post-contrast T1-weighted (T1c)

• early post-contrast T1-weighted (T1ce)

• diffusion-weighted imaging

• apparent diffusion coefficient maps

• susceptibility-weighted imaging

The four MRI archives are:

• 2_1_preop_scans_coreg.zip

• 2_2_postop_scans_coreg.zip

• 2_3_preop_scans_raw.zip

• 2_4_postop_scans_raw.zip

Corresponding acquisition and availability metadata are supplied in:

• 2_1_preop_scans_coreg_meta.csv

• 2_2_postop_scans_coreg_meta.csv

• 2_3_preop_scans_raw_meta.csv

• 2_4_postop_scans_raw_meta.csv

Images are distributed as NIfTI files within patient-ID folders.

Examination identifiers are encoded in filenames. For example:

123_0_brain_t1c.nii.gz

The second numeric field distinguishes multiple examinations belonging to the same patient.

The raw archives contain the released imaging data after de-identification and skull stripping. The co-registered archives provide the four basic sequences (T1n, T1c, T2f, and T2w) after the study preprocessing workflow.

Direct identifiers and image-file metadata were removed before release, and skull stripping was applied to reduce potentially identifying facial anatomy. These measures reduce, but cannot completely eliminate, the residual theoretical risk of re-identification inherent in sharing individual-level imaging data.

The released molecular data contain recorded somatic and clinical molecular variables rather than raw germline sequencing data.

3. Segmentations

Segmentation masks are supplied in:

• 3_1_preop_masks.zip

• 3_2_postop_masks.zip

The masks correspond spatially to the co-registered four-sequence MRI archives.

The segmentation labels are:

• segment_1 — necrotic tumour core

• segment_2 — FLAIR abnormality

• segment_3 — enhancing tumour

• segment_4 — resection bed

Initial preprocessing and segmentation were performed using the FL-PoST/FeTS workflow. The resulting masks were subsequently reviewed and manually refined by radiologists with neuro-oncologic imaging expertise.

Inter-observer reproducibility results, including Dice similarity coefficient and HD95 summaries, are provided in:

5_technical_validation.zip

4. Radiomics

Radiomics data are provided in:

• 4_radiomics_preop.csv

• 4_radiomics_postop.csv

Radiomic features were extracted from the four segmented MRI sequences and tumour subregions using the public PyRadiomics-based pipeline included in:

6_reproducibility_code.zip

The radiomics tables contain:

• first-order features

• shape features

• GLCM features

• GLRLM features

• GLSZM features

• GLDM features

• NGTDM features

• diagnostic and linkage columns

No imputation, feature normalization, or cross-scanner harmonization was applied to the released radiomics tables.

Scanner- and field-strength-association results are provided in:

5_technical_validation.zip

These analyses characterize observed scanner-associated variation and should not be interpreted as demonstrating test-retest reproducibility or scanner invariance. Users should account for scanner and acquisition heterogeneity when performing radiomics analyses.

Feature-column names follow the structure:

<image_transform_or_diagnostics>_<feature_class>_<feature_name>.<image_type>

The image types are:

• brain_t1c

• brain_t1n

• brain_t2w

• brain_t2f

Each row represents one segmentation annotation.

The following columns support linkage and quality assessment:

• identifier

• mask_label

• image_path

• mask_path

• label_sum

• error

5. Technical validation

5_technical_validation.zip contains the public artifacts generated during technical validation of the dataset, including:

• scanner-manufacturer and magnetic-field-strength association summaries

• feature-level scanner-association results

• inter-observer case-level segmentation reproducibility results

• inter-observer summary statistics

• public mask-pairing and quality-control manifest using final Case IDs

• Dice and HD95 distribution plots

• visual overlay quality-control workbook

These files are intended to make the technical validation analyses transparent and independently inspectable.

6. Reproducibility code

6_reproducibility_code.zip contains public snapshots of the code and documentation used in dataset preprocessing, radiomics extraction, quality control, and technical validation.

The archive includes:

• radiomics extraction pipeline

• batch SynthStrip wrapper

• FL-PoST documentation

• release quality-control scripts

• scanner-association analysis scripts

• inter-observer reproducibility analysis scripts

The archive contains no MRI data, patient-specific local paths, or Git metadata.