The WAW-GliomAI dataset: Multiparametric MRI, Manual Segmentations, Radiomics and Molecular Profiles of Diffuse Gliomas

The WAW-GliomAI dataset contains pre- and postresection multiparametric brain MRI images from N=192 adult patients with histopathologically confirmed diffuse gliomas (WHO grades 2–4), annotated with multicompartment tumor segmentations, comprehensive molecular profiling, and radiomics features extracted for each tumor subregion. The dataset is accompanied by clinical and demographic information.AbstractGliomas are the most common primary malignant brain tumors in adults. The latest 2021 WHO CNS classification emphasizes the role of molecular biomarkers such as IDH mutation, 1p/19q co-deletion, and MGMT promoter methylation in grading and guiding treatment decisions for gliomas. Molecular profiling, however, requires invasive sampling through biopsy or resection, which is not always feasible or risk-free.To support AI-based research approaches, we present the WAW-GliomAI dataset - an imaging-genomics resource that integrates multiparametric MRI, expert-validated tumor segmentations, radiomic features, and corresponding molecular data obtained through next-generation sequencing (NGS). The dataset contains detailed annotations for tumor subregions (enhancing tumor, necrosis, FLAIR-hyperintense abnormality and resection cavity), standardized MRI image collection, and clinically relevant molecular alterations including IDH1/2 mutation status, 1p/19q co-deletion, TERT, EGFR amplification, and more.This dataset enables research in radiogenomics, molecular prediction, and explainable AI models to support glioma classification and personalized treatment strategies, aligned with the WHO CNS 5 framework.Technical infoThe WAW-GliomAI dataset includes four major components:1. Clinical and molecular data (tabular form)All patient-level clinical and molecular variables are provided in a tabular file named “clinical__molecular_data.” Each patient is assigned a unique identifier (pat_id), which links imaging data, segmentations, and radiomics. Molecular profiling includes key biomarkers coded in binary form: IDH1/2, ATRX, TP53, TERT promoter, 1p/19q co-deletion, CDKN2A/B, EGFR, and chromosomal aberrations (e.g., 7+/10−). A glossary of variables is included in the accompanying “clinical_molecular_metadata_dictionary” file. 2. MRI imaging (NIfTI format)Pre- and postoperative MRI are avaliable and include the following sequences:T2-weighted imaging (T2w), Fluid-attenuated inversion recovery (FLAIR, T2f), Non-contrast T1-weighted imaging (T1n), Delayed post-contrast T1-weighted imaging (T1c), Early post-contrast T1-weighted imaging (T1ce), Diffusion-weighted imaging (DWI) with b-values ranging 0-1500 s/mm², Apparent diffusion coefficient (ADC) maps, Susceptibility-weighted imaging (SWI). MRI data are stored in NIfTI format. Each patient has a dedicated folder labeled with their id, containing individual files for each sequence (e.g., 123_flair.nii.gz, 123_t1ce.nii.gz). Skull stripping and irreversible anonymization were applied to all imaging data. Apart from these steps, all image sequences are provided in their raw, unprocessed state. Additionally, coregistration and bias field correction were performed for the four basic MRI sequences (T1n, T1c, T2f, and T2w). The MRI data are divided into four parts: all preoperative scans (distributed across three archives for easier download — gliomai_preop_scans_all_1.zip, gliomai_preop_scans_all_2.zip, gliomai_preop_scans_all_3.zip), all postoperative scans (gliomai_postop_scans_1.zip–3.zip), and the coregistered basic four-sequence sets for both preoperative (gliomai_preop_scans_co) and postoperative (gliomai_postop_scans_co) data.Within each archive, every patient has a separate folder labeled with their numeric id. Files are named according to the convention: for a patient with id 123 — 123_0_brain_[series name, e.g., t1c], and so on. For patients with more than one available examination, subsequent studies are numbered sequentially in the second position of the filename, for example: 123_0_..., 123_2_..., 123_3_..., and so on.Each of these archives is accompanied by a separate metadata file, named correspondingly with the suffix _metadata (e.g., gliomai_preop_scans_all_1_metadata). 3. Segmentations (NIfTI format)Image segmentations were generated for four subregions:segment_1: necrotic tumor core segment_2: FLAIR abnormalitysegment_3: enhancing tumorsegment_4: resection bedSegmentation masks are provided in two archives: gliomai_preop_masks and gliomai_postop_masks.Files are named according to the convention 1_1_tumor_mask_post.nii and correspond exactly to the coregistered basic MRI sequences. Each mask matches the respective image files from the glioma_preop_scans_co and glioma_postop_scans_co folders (these scans are co-registered, other sequences are not).In the initial stage of the annotation, image preprocessing and segmentation were performed using the Federated Tumor Segmentation (FeTS) toolkit (https://pmc.ncbi.nlm.nih.gov/articles/PMC9592188/). FeTS is an open-source framework that integrates a variety of pre-trained deep learning segmentation models and enables their application to new radiographic data. Subsequently, all segmentations were manually reviewed and refined by radiologists with expertise in neuro-oncologic imaging, in accordance with current clinical standards and best medical knowledge. 4. Radiomics (tabular form)Radiomics features were extracted from each tumor subregion using a standardized PyRadiomics pipeline. These include first-order statistics, shape descriptors, and gray-level matrix features. No imputation or normalization was performed. Radiomic characterization of WAW-GliomAI is provided in Supplementary materials 1 radiomics file. Radiomic feature data are provided in two separate files: radiomics_preop and radiomics_postop. Each line corresponds to a segmentation annotation, while each column corresponds to a feature across the different sequences. For instance, original_firstorder_Kurtosis.brain_t1c corresponds to the kurtosis first order feature in T1 with contrast, whereas original_firstorder_Kurtosis.brain_t1n corresponds to the same feature in regular T1. In essence: one row per lesion segmentation annotation.The file structure is as follows:## Output columns The output format for columns is <image_transform_or_diagnostics>_<feature_class>_<feature_name>.<image_type>. ### image_transform_or_diagnostics * diagnostics - diagnostics about the image and mask, useful for debugging* original - original image (no transforms) ### feature_class * shape - shape features* firstorder - first-order features* glcm - gray-level co-occurrence matrix features* glrlm - gray-level run-length matrix features* glszm - gray-level size-zone matrix features* gldm - gray-level dependence matrix features* ngtdm - neighborhood gray-tone difference matrix features ### image_type * brain_t1c - contrast-enhanced brain T1 image* brain_t1n - regular brain T1 image* brain_t2w - brain T2-weighted image* brain_t2f - brain T2-fluid-attenuated (FLAIR) image ### Additional columns These are specified as <column_name>.<image_type> * error - error message (if any)* identifier - identifier of the image (i.e. patient ID)* mask_label - index label of the mask* image_path - path to the image* mask_path - path to the mask* label_sum - sum of the label (size in voxels) Access privilegesThe WAW-GLIOMICS dataset is publicly available to the scientific community for non-commercial research purposes. It can be accessed via Zenodo, and use does not require co-authorship or institutional collaboration. Users are expected to cite the original publication when using the dataset.