Pituitary Adenoma Multimodal MRI (ds006248)

Researchers led from the Military University Hospital Prague shared preoperative MRI of 50 patients with pituitary adenoma on OpenNeuro. The aim is a public reference for segmenting the sellar region, where a tumor distorts the normal gland, the internal carotid arteries and the optic pathway. Each patient has a hand-made multi-class mask, and the release adds the predictions of five published segmentation methods so new models can be compared on the same cases.

Composition

There are 50 patients, 22 female and 28 male, aged 22 to 80 at surgery (mean 56.1). Every tumor in the participants file is a non-functioning adenoma; a tumor volume column exists but is empty. Each patient has up to ten series: coronal T1-weighted before and after contrast, sagittal and 3D axial contrast-enhanced T1, coronal T2, FIESTA with and without contrast, and three apparent diffusion coefficient (ADC) maps. 33 patients have ADC maps and 48 have at least one T2-type series. Diffusion appears only as ADC maps in the derivatives folder.

Acquisition

All scans come from one GE 750w 3 T scanner. Images were converted from DICOM to NIfTI in BIDS layout, and 3D scans were defaced; the defacing masks are included. The conversion, defacing and benchmarking scripts ship in the code folder.

Annotations

The first author drew masks on the coronal contrast-enhanced T1 in Brainlab Elements with five labels: tumor, normal gland, left and right internal carotid artery, and optic pathway. A second author and an external reviewer checked and refined them. Seed points for semi-automated methods and keyframe slice indices are also provided. Predictions from two nnU-Net models, a CNN ensemble, a keyframe-only CNN and the semi-automated GrowCut method are aligned to the contrast-enhanced T1 space, with Dice, IoU and Hausdorff results reported in the README.

Known limitations

Single center and single scanner. Only non-functioning adenomas are included, so hormone-secreting tumors are absent. Sequence coverage varies between patients. Ages are age at surgery. The license statements disagree: the README names CC BY-NC 4.0 and the dataset metadata names CC0, and this entry records the non-commercial license. The data descriptor paper was under review at release and is not cited here.