Liver segmentation on CT without physical geometry

PLC-CECT, delayed-phase liver segmentation

The delayed contrast phase of a primary liver cancer CT examination, with the liver label refined by clinicians: the fixed images of LiverPhasesCT, referenced on their native grid without copying or resampling.

27 test casesabdomenimage CT

Look at the images

Read before comparing (3)

Split. LiverPhasesCT's partition is kept: 27 test patients and 24 distinct development patients. Only the delayed fixed image is used.
Images. The provider's images lost their physical voxel size and hold an 8-bit soft-tissue window, not Hounsfield units. As LiverPhasesCT, the dataset is ranked apart, and its surface distances are not millimetres.
Labels. Only the liver is scored; the parent's separate lesion label set does not stand in for it.

		Accuracy			Cost
#	Method	Dice	Surface Dice	HD95 mm	Time s	RAM GB	GPU GB	Images
11–2	KonfAI segmentation app · TotalSegmentator (total), KonfAI appGPUapp	0.716 ±0.089	0.421 ±0.064	47.35 ±10.37	26	3.6	6.3
21–2	KonfAI segmentation app · TotalSegmentator (total), the authors' packageGPUapp	0.668 ±0.104	0.407 ±0.076	58.30 ±11.77		16.3	3.9
33–4	KonfAI segmentation app · MRSegmentator, KonfAI appGPUapp11 failed	0.406 ±0.147	0.149 ±0.066	58.32 ±17.68	44	19.8	6.6
43–4	CT-FM whole-body segmentationGPUapp2 failed	0.320 ±0.083	0.208 ±0.043	69.82 ±15.15	57	13.4	7.1
5	MONAI wholeBody CTGPUapp7 failed	0.061 ±0.043	0.019 ±0.014	157.10 ±29.26	17	4.1	8.1

The order averages each method's ranks on Dice, Surface Dice, HD95 mm, so the first one is not always the best in every column. ± is the 95% interval of the mean over the 27 cases. Two neighbours are tied when a paired Wilcoxon test on the shared cases does not find the upper one better on Dice (5%): the places a method could hold are then shown under its own. The best value of each column is in colour. Click a row to see its images, a column title to sort by it. Time and memory measured on Workstation.

2 results not ranked: an older run, cases still missing, or a failure on every case

		Accuracy			Cost
#	Method	Dice	Surface Dice	HD95 mm	Time s	RAM GB	GPU GB	Machine	Images
	KonfAI segmentation app · TotalSegmentator (total), KonfAI appGPUappolder run	0.716 ±0.089	0.421 ±0.064	47.35 ±10.37	16	3.1	10.6	Laptop
	MONAI wholeBody CTGPUappolder run	0.134 ±0.057	0.042 ±0.018	138.85 ±24.30	11	2.7	9.3	Laptop

Dice: how much the organ contours overlap, 1 is perfect. Surface Dice: the share of the contours' surfaces within a tolerance of each other, 1 is perfect. HD95: distance between the contours' surfaces.

Case by case

Measure

Dice	1	2	3	4	5	6	7	8	9	10	11	12	13	14	15	16	17	18	19	20	21	22	23	24	25	26	27
KonfAI segmentation app · TotalSegmentator (total), KonfAI app	0.846	0.765	0.784	0.622	0.740	0.830	0.816	0.876	0.802	0.683	0.874	0.183	0.835	0.835	0.788	0.874	0.899	0.933	0.783	0.838	0.569	0.802	0.787	0.784	0.046	0.276	0.463
KonfAI segmentation app · TotalSegmentator (total), the authors' package	0.812	0.862	0.731	0.159	0.399	0.750	0.686	0.850	0.368	0.568	0.844	0.090	0.846	0.911	0.868	0.594	0.914	0.917	0.873	0.838	0.604	0.768	0.896	0.869	0.059	0.345	0.604
KonfAI segmentation app · MRSegmentator, KonfAI app	0.793	0.811	0.814	failed	failed	failed	0.244	0.712	0.784	0.637	0.740	0.475	0.833	0.676	failed	0.824	0.847	0.881	failed	failed	failed	failed	failed	failed	0.532	0.350	failed
CT-FM whole-body segmentation	0.590	0.370	0.240	failed	0.133	0.247	0.550	0.348	failed	0.178	0.356	0.269	0.340	0.305	0.775	0.323	0.619	0.525	0.662	0.288	0.159	0.212	0.184	0.604	0.000	0.164	0.203
MONAI wholeBody CT	0.002	0.003	0.204	0.191	0.001	0.022	0.000	failed	0.000	failed	failed	0.000	failed	0.367	0.090	failed	failed	failed	0.018	0.398	0.029	0.117	0.090	0.086	0.029	0.004	0.000

Click a score to look at that result: the image, the truth in green and the method's map in red. Underlined scores have their image; for the others you see the case without the result.

About the data

The first test caseClick to look at every case.

Truth: contours by experts the liver (label 1) on each phase: the LiverMask network's segmentation refined by clinicians in ITK-SNAP
Source: doi.org/10.57760/sciencedb.12207
doi.org/10.1038/s41597-025-05125-2
github.com/ljwa2323/PLC_CECT
Licence: CC BY 4.0

24 cases for tuning, never tested

P0001/fixed, P0095/fixed, P0194/fixed, P0165/fixed, P0162/fixed, P0115/fixed, P0171/fixed, P0024/fixed, P0113/fixed, P0119/fixed, P0155/fixed, P0033/fixed, P0111/fixed, P0075/fixed, P0062/fixed, P0088/fixed, P0035/fixed, P0266/fixed, P0041/fixed, P0217/fixed, P0272/fixed, P0228/fixed, P0203/fixed, P0246/fixed