Brilliance big bore scanner

To evaluate the robustness of the final deep learning segmentation network, a dataset of over 15,000 deceased subjects with different acquisition parameters and population distribution was acquired from the NMDID.³⁴ We selected 10 patients with death by natural cause, age under 50 and similar positioning to patients from the UMCU dataset: 6 male and 4 female patients, with a mean age of 34 years (SD: 6.4 years). For each subject, two CT scans were available which included the lower extremities. A torso scan, including the coxae and femur, and a lower leg scan, including the tibia, fibula, talus, calcaneus and femur. All CT scans were acquired on a Philips Brilliance Big Bore scanner (Philips Medical Systems). The CT acquisition parameters were: tube voltage = 120 kVp, tube current = 82–245 mA, effective dose = 100–301 mAs, slice thickness = 1 mm, slice increment = 0.5 mm, pixel spacing = 0.63–1.17 mm, matrix size = 512 × 512 pixels. The images were resampled using trilinear interpolation to isotropic 0.8 × 0.8 × 0.8 mm to match the voxel size of the UMCU dataset more closely. Due to high noise levels in the proximal part of the lower leg scan and in the torso scan, a Gaussian filter (sigma = 0.5, kernel size = 4) was applied before segmentation. Examples of CT scans from both datasets are shown in Figure 1.