nnUNetv1 model trained for ~1 month on a single A6000 GPU using 1,578 CT volumes from VerSe and TotalSegmentator. As VerSe does not label vertebrae that appear partially in the image, we crop out the inferior-most 20 pixels and superior-most 20 pixels of all VerSe scans. All volumes are resampled to 1.5mm spacing in all dimensions, reoriented to RAS+, and the mirroring training augmentation is disabled, following the strategy of TotalSegmentator. Label values are as follows:
{
"0": "background",
"1": "C1",
"10": "T3",
"11": "T4",
"12": "T5",
"13": "T6",
"14": "T7",
"15": "T8",
"16": "T9",
"17": "T10",
"18": "T11",
"19": "T12",
"2": "C2",
"20": "L1",
"21": "L2",
"22": "L3",
"23": "L4",
"24": "L5",
"3": "C3",
"4": "C4",
"5": "C5",
"6": "C6",
"7": "C7",
"8": "T1",
"9": "T2"
}
