This data set consists of 392 high resolution (1.5mm axial slice thickness) GammaKnife MR/CT pairs collected concurrently. Using the SRS frame, the T1-MR images were rigidly registered to the corresponding CTs. The data set is approximately equal parts healthy patients (trigeminal neuralgia or acoustic neuroma) and patients with metastatic disease. Cases were excluded for anatomic altering lesions (i.e. meningiomas) or tumors nearby the hippocampus. Contours of the left and right hippocampus were created by the researchers following the RTOG 0933 protocol and a head contour was created to be used to mask out the SRS frame.

The dataset also includes:

• MR Sequences: Pregadolinium (Sagittal T1, axial multiecho, axial FLAIR, axial MPR) or Postgadolinium (Axial T1, axial MPR) or T2 MRI for benign cases
• Clinical data: Tumor primaries or benign diagnosis (TGN, Acoustic neuroma)
• GammaKnife treatment planning contours. Although we just upgraded our GammaKnife system so I am not certain all will be available.
• Deep learning generated contours of the hippocampus from this publication: https://doi.org/10.1002/mp.14098

The high-resolution MR/CT image pairs allows researchers to investigate segmentation of the hippocampus, and to contour additional structures to expand upon the availability of cranial segmentation data.

Image AddedThis collection is comprised of 390 patients who presented with vestibular schwannoma (VS, n=73), trigeminal neuralgia (TGN, n=119) or metastatic disease (M, n=198, 4-26 metastases) and were subsequently treated with Gamma Knife (Elekta AB, Stockholm, Sweden) stereotactic radiosurgery. For each patient, the treatment indication is designated with a suffix on their patient ID (VS, TGN or M).

All patients are provided with at least one high-resolution (1 mm slice thickness) T1 FLASH trans-axial MR imaging study and their corresponding high-resolution axial planning CT. When available, treatment planning data (struct, dose, plan), alternative MR sequences (FLAIR, T2 CISS, etc.) and follow-up MR imaging studies were collected. Each MR image used during treatment planning was registered to the CT frame of reference and is provided with the DICOM registration file and the aligned secondary image. Additionally, for each patient in the cohort, hippocampal contours generated by multiple institutional observers are provided in a separate structure set. In total, this dataset contains: CT (n=390), MR (n=3868), REG (n=872), DOSE (n=928), PLAN (n=928), planning STRUCT (n=931), and hippocampal research STRUCT (n=390).

Planning Data

Gamma Knife planning studies are designated by “Gamma Knife Planning Data” in the study description DICOM header tag. For each patient, MIM (MIM Software, Beachwood, OH) was used to generate a rigid registration between CT and each MR sequence and the registration accuracy was validated using the stereotactic frame fiducial markers. From each registration, a DICOM RTREG file and aligned secondary are provided with each aligned secondaries series indicated by “[original series description] Co-registered to CT” in the series description DICOM header tag. During export from GammaPlan (Elekta AB, Stockholm, Sweden), the treatment planning files are provided in duplicate for each imaging modality frame of reference (CT and each MR sequence). Treatment planning structure names were renamed to remove misspellings, synonyms, and identifiable information. Structures with the format Plan1[tgt#]#Gy denote each target volume (a, b, ...) and the dose to the specified target in Gray. For patients with multiple lesions, ROI structure names are given as the brain region followed by the relative anatomical directions (e.g. Lobe_Frontal_R, Lobe_Parietal_L). Any empty contours were preserved but are designated with '(Empty)' in the name.

Follow-up Data

When available, patient follow-up images were collected between the provided treatment date and either the next Gamma Knife treatment or up-to approximately two years following treatment. In total, 197 patients are provided with follow-up imaging studies, with a median of 2 (range 1-13) follow-up studies provided per patient. Each follow-up imaging study is designated as “Follow-up Image Set #” in the Study Description. Each follow-up study exists in a unique frame of reference and was not co-registered to the original treatment planning CT volume.

Hippocampal Contours

Three independent observers generated hippocampal contours from the CT aligned-secondary of the T1-weighted MR image, with the resultant contours saved to the CT frame of reference. In total, 744 unique left, right contour pairs were generated (observer 1, n=390; observer 2, n=247; observer 3, n=107). In addition to hippocampal contours, the region grow tool was used to generate a head contour (ROI name ‘head’) to mask out the stereotactic frame and remove most of the reconstruction artifacts on the inferior extent of the image volume.

Additional Information

All MR images designated as ‘Co-registered to CT’ in the Series Description were interpolated to the CT frame of reference, potentially altering the field of view or voxel resolution. For all co-registered MR images, the original primary images and registration files are provided for use as needed.