A Multi-Patient Brain CT Dataset for Edge-AI-Enabled Intracranial Hemorrhage Detection on Raspberry Pi

Abstract

This dataset comprises 852 multi-angle brain Computed Tomography (CT) scan images collected across multiple patients presenting with intracranial hemorrhage (ICH), structured to support rapid computer-aided diagnosis and edge computing research. All images are standardized in PNG format and indexed by individual patient cases (e.g., sequential cross-sectional slices per patient) to facilitate robust, patient-level train-validation-test partitioning without data leakage. Bounding-box annotations are provided in standard YOLO format to enable direct training, evaluation, and deployment of lightweight object detection architectures—specifically the nano variants of YOLO—on resource-constrained single-board computers such as the Raspberry Pi. The dataset is fully anonymized, with all Protected Health Information (PHI) and DICOM metadata removed, providing a reliable benchmark for developing low-latency Internet of Medical Things (IoMT) screening tools, mobile triage systems, and automated point-of-care neuroimaging diagnostics in resource-limited clinical environments.