license: apache-2.0 | |
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" | |
} |