Synchrotron X-ray Data of non-diseased and diabetic Human Pancreas Tissue and Islet Segmentation Model
文献摘要
Three-dimensional virtual histology using synchrotron radiation micro-computed tomography enables non-destructive visualization of biological tissue microarchitecture with isotropic resolution. To support quantitative studies of human pancreatic morphology, we provide high-resolution 3D X-ray imaging datasets of 55 pancreatic tissue cores from 32 individuals. The collection comprises histologically non-diseased pancreatic tissue from organ donors and non-neoplastic regions of surgical resection specimens from patients with pancreatic cancer, including subgroups with insulin-dependent diabetes or following neoadjuvant chemotherapy. Formalin-fixed, paraffin-embedded tissue cores with a diameter of 1 mm were imaged at 650 nm isotropic voxel size, yielding a mean effective spatial resolution of 1.66 µm. The cohort includes reconstructed volumes, corresponding immunohistochemistry, clinical metadata, expert-validated annotations, and dataset-specific example 2.5D U-Net models for automated segmentation of islets of Langerhans. A representative co-registered section yielded a Dice similarity coefficient of 0.81 and intersection-over-union of 0.67. This openly accessible, FAIR-compliant resource supports quantitative analysis of endocrine, exocrine, ductal, and stromal architecture and provides a reproducible reference framework for three-dimensional virtual histology of human pancreatic tissue and computational benchmarking.