Spinal-Multiple-Myeloma-SEG dataset expands with 72 dual-energy CT scans

A team led by Michal Nohel has published an extension to the Spinal-Multiple-Myeloma-SEG dataset, a dual-energy CT resource designed for multiple myeloma research. The new version adds expert-validated voxel-wise segmentations of the trabecular compartment in lumbar vertebrae, enabling deeper analysis of bone microstructure. It includes 72 dual-energy CT examinations from 67 adult patients (mean age 66, 36% female), acquired retrospectively using a dual-layer dual-energy CT system. Imaging data covers conventional CT, virtual monoenergetic images, and calcium-suppressed reconstructions, alongside structured clinical metadata.

The segmentation masks were initially generated using a pretrained nnU-Net model, then refined through manual expert correction and radiological quality control to ensure anatomical consistency. All imaging data is provided in DICOM format, while masks are available in NIfTI and DICOM-SEG. The original dataset remains accessible via TCIA, while the masks (Version 2) are immediately available on Zenodo under CC BY 4.0, with eventual integration into TCIA after curation. The dataset is intended for trabecular bone segmentation, bone mineral density analysis, and deep-learning model development for vertebral anatomy in multiple myeloma, supporting both segmentation and multimodal learning tasks.