AortaSeg-60: An Open Real-World CT-Angiography Dataset of the Aorta with Automated Segmentation Masks and Pathological Variability

We present AortaSeg-60, an open dataset of 60 real-world thoraco-abdominal CT-angiography scans of the aorta encompassing normal anatomy and pathological variations, designed for AI research, benchmarking, and educational purposes. The dataset is organized into six balanced categories: young normal, elderly normal, aortic aneurysms, aortic dissections, venous acquisition, and non-contrast acquisition, capturing realistic anatomical and pathological diversity. All scans are provided in NIfTI format with fully automated aortic segmentation masks generated using TotalSegmentator, without manual correction, enabling evaluation of typical algorithmic errors and testing of refinement strategies. Two radiologists performed a technical validation to ensure dataset curation and correct category assignment. AortaSeg-60 is publicly available on Zenodo under a CC0 license. By providing paired imaging and automated labels, the dataset facilitates reproducible research, algorithm development, and method comparison for vascular segmentation, while noting limitations of sample size, single-centre acquisition, and reliance on automated annotations.