Answers are grounded in this paper and cite their sources.
Publicly available traumatic brain injury (TBI) imaging datasets remain limited by the absence of longitudinal imaging, standardized annotations, and linked clinical outcomes. We developed and validated the Shiraz-TBI (STBI) dataset, a large-scale multimodal resource comprising 3602 patients, 12,155 serial non-contrast brain CT examinations, and 376,195 DICOM images, with all studies additionally available in NIfTI format. Radiologists generated structured annotations for intracranial hemorrhage subtypes, mass effect, midline shift, skull fractures, Marshall CT scores, and other imaging features, while temporal variables and time-to-surgery outcomes were derived. Internal validation by neurosurgeons demonstrated near-perfect inter-rater agreement (κ > 0.85 for most variables). A multimodal subset of 1259 patients includes demographic, clinical, laboratory, treatment, and functional outcome data. By integrating patient-level longitudinal chronologically ordered serial neuroimaging with standardized annotations, temporal information, and prognostic outcomes, STBI provides a publicly available, machine learning-ready resource for imaging analysis, disease progression modeling, prognostic prediction, and clinical decision support in TBI.
No takes yet. Share an insight, caveat, or question.
Andishgar et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: