universitytehran/MRI-GBM-MET
MRI-GBM-MET: Brain MRI Dataset (GBM vs. MET) Dataset Summary The MRI-GBM-MET dataset is a collection of axial brain MRI scans from 90 human subjects diagnosed with either Glioblastoma (GBM) or Brain Metastasis (MET). It is designed to support medical image classification, diagnostic research, and multimodal evaluation tasks. All subject-level diagnostic labels are histopathologically confirmed (post-surgical biopsy or resection) and are completely independent of… See the full description on the dataset page: https://huggingface.co/datasets/universitytehran/MRI-GBM-MET.
MRI-GBM-MET: Brain MRI Dataset (GBM vs. MET)
Dataset Summary
The MRI-GBM-MET dataset is a collection of axial brain MRI scans from 90 human subjects diagnosed with either Glioblastoma (GBM) or Brain Metastasis (MET). It is designed to support medical image classification, diagnostic research, and multimodal evaluation tasks.
All subject-level diagnostic labels are histopathologically confirmed (post-surgical biopsy or resection) and are completely independent of radiological masks. The dataset includes pixel-level expert annotations from board-certified neuroradiologists used to identify lesion-containing slices.
Dataset Configurations
For each subject and modality (T1CE and T2), four distinct configurations are provided:
Key Statistics
- Total Subjects: 90 (45 Glioblastoma, 45 Brain Metastasis)
- Total Slices: 4,091 images in
.jpgformat - T1-weighted Contrast-Enhanced (T1CE) MRI: 2,199 slices
- T2-weighted MRI: 1,892 slices
- Sequence Length per Subject: Median of 20 slices (range 15–160)
- ROI Slices per Subject:
- T1CE: Median of 8 slices
- T2: Median of 11 slices
Technical Specifications & Metadata
- Format: All images are in
.jpgformat. - Naming Convention: Files are named as
slice_{index}.jpgwhere{index}represents the absolute global sequence position within the original scan. - Consistency: Filenames are aligned across configurations. If a lesion is located on slice 10, the corresponding image
slice_009.jpgis present inFullVolume,ROISlices, andROICrops. - Integrity Rule: For any subject sequence: $$\text{Total Slices (FullVolume)} = \text{ROISlices count} + \text{NonROI count}$$
