Introduction:

Imaging data:

Preoperative MRI comprises four structural MRI sequences: native T1-weighted (T1), contrast-enhanced T1-weighted (ceT1), native T2-weighted (T2), and T2-weighted fluid-attenuated inversion recovery (T2-FLAIR). These scans were acquired before surgery using various scanners at multiple institutions, making their acquisition parameters heterogeneous. Unlike preoperative MRI, all intraoperative MRI were acquired in the AMIGO suite at the Brigham and Women’s Hospital (Boston, USA) using a 3T wide-bore (70 cm) MRI scanner. All iUS series were acquired using a tracked 2D neuro-cranial curvilinear transducer. The transducer was swept unidirectionally through the craniotomy at a slow, consistent speed. This specific motion, in conjunction with the tracking, enabled the reconstruction of a 3D volume from the tracked 2D sweeps.

Segmentation data:

Various segmentations were created to assist the surgical resection. These include manual segmentations of the preoperative whole tumor, preoperative tumor target (i.e., the radiologically identifiable tumor specifically targeted for resection), resection cavity resulting from prior surgery (i.e., in case of reoperation), intraoperative residual tumor, and the automatic segmentations of cerebrum and ventricles (Brainlab AG, Munich, Germany). Only structures deemed necessary for the surgical resection by the attending neurosurgeon were segmented. Specifically, segmentations of the manual preoperative whole tumor (113 cases), preoperative tumor target segmentations (3 cases), manual previous resection cavity segmentations (21 cases), residual tumor segmentations (58 cases), and automated segmentations of the cerebrum (89 cases) and ventricles (54 cases). All cerebrum, ventricle, and tumor segmentations were created preoperatively during the surgical planning stage. In contrast, residual tumor segmentations were created intraoperatively from iMRI.

Clinical metadata:

Demographic information, including age, sex, and ethnicity, was obtained from the corresponding patient medical records. The age range of the included population was 20–76. The ratio of male:female was equal to 61:53. Moreover, clinico-pathologic data such as the tumor type, tumor grade, radiological characteristics upon contrast administration, tumor location, and the reoperation status were assessed by the treating neurosurgeons. Tumor type and grade were specified according to the World Health Organization (WHO) 2021 Classification of Tumors of the Central Nervous System. Additionally, tumors were classified into one of 3 categories based on proximity to the functional cortex (non-eloquent, near eloquent, and eloquent). 

Pre-processing:

The MRI and ultrasound images are provided in DICOM format. All MRI images were defaced using automatic affine registration or manual landmark registration with NiftyReg and the template and face mask provided in pydeface. The code of the algorithm is publicly available.