Signa premier

All participants underwent an MRI session during which three-dimensional (3D) T1-weighted and resting-state functional MRI (RS-fMRI) T2*-weighted sequences were acquired. The data were collected using a 3T MRI (Signa^TM Premier, General Electric Company, Milwaukee, WI, USA) equipped with a 48-channel coil and installed at the Cliniques Universitaires Saint-Luc (UCLouvain, Brussels, Belgium). The 3D T1 encompassing the whole brain was selected to provide detailed anatomy (1 mm³) due to an MPRAGE sequence (inversion time = 900 ms; repetition time (TR) = 2188.16 ms; echo time (TE) = 2.96 ms; flip angle (FA) = 8°; field of view (FOV) = 256 × 256 mm²; matrix size = 256 × 256; 156 slices; slice thickness = 1 mm; no gap; and total scan time = 5 min 36 s). The RS-fMRI sequences were collected with hyperband (Factor 3) echo-planar imaging: FOV= 220 × 220 mm²; matrix size = 110 × 110; TE = 30 ms; FA = 90°; slice order ascending and interleaved; slice thickness = 2 mm; and ARC 2 (parallel imaging). The TR was 1500 ms and the number of slices was 64, with the whole brain scanned 250 times per run (=6 min 15 s).

All MRI scans were acquired on a 3 T scanner (Discovery MR 750 or Signa Premier, GE Medical Systems, WI, USA), using an 8-channel coil^{34 (link),54 (link),55 (link)}. Anatomical and perfusion images (ASL) were acquired in all cases. Parameters for anatomical image acquisition have been provided in our previous work^{53 (link)}, while those for ASL acquisition were: pseudo-continuous ASL, echo time 10.5–11.2 ms, repetition time 4531–4742 ms, single post labelling delay of 1525 ms, flip angle 111°, averages 3, slice thickness 4 mm, spacing between slices 4 mm, and spiral readout. Quantitative perfusion maps were automatically generated by GE reconstruction software. Children underwent sedation if clinically indicated^{26 (link)}. Since no clinical seizures were noted during the MRI scan, the perfusion patterns shown by the scans were considered to mirror the brain perfusion during the interictal state.

Height (cm) and body weight (kg) were measured using a stadiometer (YHS‐200D, YAGAMI Inc., Nagoya, Japan) and an anthropometer (MC‐980A, Tanita, Tokyo, Japan), respectively. Body weight was measured with light clothing and without shoes. BMI (kg/m²) was calculated from height and body weight measurements. Body fat (%) was measured using bioelectrical impedance analysis (MC‐980A, Tanita, Tokyo, Japan). Fat‐free mass (kg) was calculated from the body weight and fat. Visceral fat area (cm²) and subcutaneous fat area (cm²) were measured using MRI (Signa Premier; GE Healthcare, Waukesha, WI, USA), as described previously (Usui et al., 2020 (link)). Body composition at each body part was measured using DXA, as described in a previous study (Kawakami et al., 2021 (link)). Abdominal circumference (cm) was measured to the nearest 0.1 cm at the umbilical region with an inelastic measuring tape at the end of normal expiration. Calf circumference (cm) was measured in 0.1 cm increments in the standing position; twice on each side where the circumference was the greatest. Details of the two circumferential measurements can be found in previous studies (Kawakami et al., 2020 (link)).

The scans in the in-house dataset were obtained at nine institutions with 1.5 T (Avanto, Avanto-fit, and Aera, Siemens Healthcare, Erlangen/Germany; Signa HDxt Signa, General Electric Healthcare, Chicago/USA) or 3 T scanners (Prisma, Skyra, and Vida, Siemens Healthcare, Erlangen/Germany; Signa Premier, GE Healthcare, Chicago/USA).
All mpMRI or bi-parametric MRI protocols followed PI-RADS version 2 or 2.1. At a minimum, the bi-parametric prostate MRI protocol encompassed tri-planar T2-weighted and diffusion-weighted imaging. The diffusion-weighted imaging was performed with echo-planar imaging in axial planes with at least three b-values. Some patients had an acquired DWI with a b-value ≥ 1400 s/mm², while others had calculated DWI with a b-value of 1400 s/mm² following the PI-RADS. The ADC maps were calculated using a linear least-square fitting with all acquired b-value. We did not use dynamic contrast-enhanced images since the challenge organizers did not provide them. Further details of the MRI protocols were omitted for the sake of brevity.

Images were obtained on a 3.0 T magnetic resonance scanner (Signa Premier, GE Healthcare; Chicago, Illinois, USA) through a 48-channel head coil at the National Neurological Center in Beijing Tiantan Hospital. Foam padding and earplugs were used to limit head movement and reduce the effect of noise on participants. All participants were instructed to hold a supine position, keep their eyes closed, and remain awake during the scan. The BOLD acquisition time was 330 s obtained using a multiband BOLD sequence with a 2.4 × 2.4 × 2.4 mm³ voxel size. The parameters were as follows: time point = 330, slice number = 65, flip angle = 64°, repetition time (TR) = 2000 ms, echo time (TE) = 68 ms, and field of view (FOV) = 208 mm × 208 mm. A high-resolution 3D T1-weighted structural image was acquired using the MP-RAGE sequence with 1.0 mm³ isotropic voxels. Its parameters were slice number = 192, flip angle = 8°, preparation time = 880 ms, recovery time = 400 ms, acquisition time = 240 ms, and FOV = 250 mm × 250 mm.

3T MR imaging (SIGNA Premier; GE Healthcare) was performed within 4 weeks before or after DSA on all 53 patients. Patients did not receive microsurgery, endovascular treatment, or radiotherapy between MR imaging and DSA.
The conventional MR imaging protocol included axial T2WI (flip angle 8°; TR 4924 ms; TE 107.7 ms; section thickness 5.0 mm; NEX 1; bandwidth 62.50 kHz) and 3D sagittal T1 MPRAGE (flip angle 8°; TR 1000 ms; TE 2.7 ms; section thickness 1.0 mm; NEX 1; bandwidth 31.25 kHz).
The parameters of silent MRA were as follows: field of view (FOV), 20 × 20 cm; matrix, 150 × 150; flip angle, 5°; TR, 828 ms; TE, 0 ms; section thickness, 1.2 mm; NEX, 1; bandwidth, 25 kHz; and acquisition time, 6 min 59 s; and the labeling duration, 2034 ms. 3D radial sampling was applied during the readout scheme. A 48-channel head coil was used for all patients. To avoid the influence of the labeled position on the results, the lower edge of the FOV was located at the lower edge of the C2 vertebral body level in each patient.

All imaging was performed on a 3T MR system (Signa Premier, GE Healthcare, Milwaukee, WI) using a 48‐channel head coil from the same vendor. Motion estimates for prospective correction were obtained with a markerless motion tracker (Tracoline TCL3.1m, research version provided by TracInnovations, Ballerup, Denmark), including the software TracSuite v3.0m. Both the hardware and software of the tracking system were customized to fit the current research project. Acquisitions were made in vivo in accordance with the institutional review board policy, and informed consent was obtained from both volunteers.