Automatic segmentation of white matter lesions on multi-parametric MRI: convolutional neural network versus vision transformer

Table 6 Performance evaluation of 3D models in the 2017 MICCAI WMH Segmentation Challenge

	DSC	H95(mm)	AVD(%)	Lesion Recall	Lesion F1
bigrbrain	0.77	9.46	28.04	0.78	0.71
cian	0.78	6.82	21.72	0.83	0.70
himinn	0.62	24.49	44.19	0.33	0.36
misp	0.78	11.10	19.71	0.68	0.71
neuro.ml	0.78	6.33	30.63	0.82	0.73
achilles	0.63	11.82	24.41	0.45	0.52
tignet	0.59	21.58	86.22	0.46	0.45
upc_dlmi	0.53	27.01	208.49	0.57	0.42
nic-vicorob	0.77	8.28	28.54	0.75	0.71
nus_mnndl	0.76	6.92	50.28	0.88	0.71
Proposed CNN-based model	0.77	5.36	18.72	0.61	0.7
Proposed Transformer-based model	0.79	3.71	20.47	0.77	0.75

Note: For each metric, the table displays the average value. Results in bold indicate the best score for each metric
Abbreviations: AVD, Average Volume Difference; DSC, Dice Similarity Coefficient; H95, 95th percentile modified Hausdorff Distance

ISSN: 1471-2377