A Paired Head CT-MRI Dateset for Cross-Modality Image Synthesis

Magnetic resonance imaging (MRI)-guided radiation therapy planning (MRIgRTP) and surgical planning (MRIgSP) enable patients to avoid computed tomography (CT) scans, thereby reducing the cumulative health risks associated with ionizing radiation exposure. However, MRI does not provide electron density information or detailed visualization of bony structures, posing significant challenges for both MRIgRTP and MRIgSP. Synthetic CT (sCT) generation from MRI has emerged as a promising solution to address these issues. However, high-quality synthesized images depend heavily on the quality and size of the training datasets, which are limited in open-access imaging database. Here, we introduce a publicly accessible, paired head CT-MRI dataset. This dataset comprises paired, thin-slice CT images (reconstructed by both soft and sharp kernels) and MRI images from 120 preoperative neurosurgical patients diagnosed with pituitary tumors, meningiomas, vestibular schwannomas, or cerebrovascular diseases. By releasing this dataset, we aim to facilitate the development and evaluation of computational algorithms for MRI-to-CT cross-modality translation, ultimately enabling cost-effective, high-quality care of brain tumor patients undergoing MRIgRTP and MRIgSP.