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 multisequence brain MRI dataset collected retrospectively at the Anti-Cancer Center of Tlemcen, Algeria. The dataset includes anonymized MRI examinations from 45 patients with primary and secondary brain tumors, including 28 primary brain tumor cases and 17 secondary brain metastasis cases. For each patient, the dataset provides multisequence MRI data, including T1-weighted, T2-weighted, FLAIR, and post-contrast T1-weighted images. The imaging data are organized at the patient level and are accompanied by a structured patient-level metadata file in CSV format. The metadata include anonymized demographic information, tumor origin, general tumor type, histological information when available, localization, laterality, metastatic source for secondary tumors, contrast enhancement pattern, necrosis, edema, multiplicity, and border characteristics. The dataset contains 19,009 MRI slices and is intended for research on primary versus secondary brain tumor classification, transfer learning, external validation, multimodal image-tabular modeling, radiomics, and weakly supervised learning. Since the dataset contains multiple slices and sequences per patient, all machine learning experiments should use patient-level splitting to avoid data leakage. Slice-level random splitting is not recommended. The current release does not include voxel-level segmentation masks. Therefore, the dataset is primarily designed for classification-oriented studies, patient-level modeling, multimodal learning, and external validation rather than fully supervised tumor segmentation. All data were anonymized prior to release. Direct identifiers were removed, and patient identifiers were replaced by study-specific anonymized codes. Users must not attempt to re-identify patients and should cite the dataset appropriately in any resulting publication.