Early Brain Disease Identification Using Transformer Model on MRI Scan Images

The dataset unites 4,601 cleaned MRI scans of five clinically relevant conditions, which provides a well-balanced and diversified basis of computer vision and medical AI experiments. The dataset contains five categories following the cleaning and scaling, namely HallervodenSpatz Disease (currently known as Pantothenate KinaseAssociated Neurodegeneration) with 972 images, Magnetic Resonance (MR) Brain with 958 images, Moyamoya Disease with Intraventricular Hemorrhage with 753 images, Neurofibromatosis Type 1 (NF1) with Optic Glioma with 838 images, and Retinoblastoma with Intracranial Spread Preprocessing was done to standardize all the images to the same resolution and to be compatible with contemporary deep-learning platforms.

To facilitate strong training, the dataset is augmented with a training split of 10000 images whereas the sets of validation (458 images) and test (464 images) do not include any modifications to ensure the integrity of model evaluation. Each of the classes is stored as different directories in a structure that is completely compatible with PyTorch, Tensorflow ImageFolder, as well as other machine-learning pipelines. The data will be used in medical image classification, feature extraction, disease recognition and the transformer-based modeling.