Tlemcen-NeuroOncMRI: A Clinically Annotated Multisequence Brain MRI Dataset with Patient-Level Metadata for Primary and Secondary Tumor Classification from Algeria
Description
Tlemcen-NeuroOncMRI is a clinically annotated, patient-linked brain MRI dataset collected retrospectively at the Anti-Cancer Center of Tlemcen, Algeria. It includes 45 independent patients, with 28 primary brain tumor (PBT) cases and 17 cerebral metastasis (CM) cases, for a total of 19,009 MRI images. The resource also provides patient-level metadata and one de-identified French-language clinical imaging report per patient. MRI examinations were acquired at multiple external imaging centres, with heterogeneous acquisition protocols and sequence availability. A clinically curated subset of 2,393 images (1,334 PBT and 1,059 CM) was used for the companion patient-level benchmark. Because multiple images originate from the same patient, all machine-learning experiments should use patient-level splitting to avoid data leakage; random slice-level splitting is not recommended. The dataset is intended for brain tumor classification, transfer learning, multimodal MRI-report modeling, radiomics, weakly supervised learning, and reproducibility studies. It does not include voxel-level segmentation masks. All data were anonymized before release, and users must not attempt to re-identify participants.