Table 2 F1, recall, and precision metrics are reported for two intersection over union thresholds, 0.5 and 0.7.

From: A review and comparison of breast tumor cell nuclei segmentation performances using deep convolutional neural networks

Segmentation Methods	Aggregated Jaccard Index	Mean Avg. Precision	F1(0.7)	Recall(0.7)	Precision(0.7)	F1(0.5)	Recall(0.5)	Precision(0.5)
(a) Nuclei segmentation results for the MoNuSeg test dataset
Otsu	0.0456	0.0677	0.0310	0.0255	0.0396	0.1619	0.1331	0.2065
Watershed	0.0828	0.1581	0.0863	0.0591	0.1594	0.2743	0.1880	0.5070
Fiji	0.3396	0.2370	0.1828	0.1447	0.2481	0.4411	0.3493	0.5986
U-Net(VGG-16)	0.4925	0.2736	0.2886	0.2927	0.2845	0.6511	0.6604	0.6420
U-Net(VGG-19)	0.4841	0.3007	0.3452	0.3480	0.3426	0.6735	0.6788	0.6683
U-Net(ResNet-50)	0.4882	0.3163	0.3772	0.3884	0.3667	0.6967	0.7173	0.6772
U-Net(ResNet-101)	0.4687	0.3318	0.3242	0.2834	0.3788	0.6133	0.5360	0.7166
U-Net(ResNet-152)	0.4396	0.3119	0.3368	0.3241	0.3506	0.6706	0.6452	0.6980
U-Net(DenseNet-121)	0.4668	0.2988	0.3579	0.3738	0.3432	0.6796	0.7099	0.6517
U-Net(DenseNet-201)	0.5083	0.3185	0.3760	0.3884	0.3645	0.6980	0.7208	0.6765
U-Net(Inception-v3)	0.4440	0.2879	0.3044	0.3005	0.3085	0.6422	0.6339	0.6507
Mask R-CNN	0.5282	0.3884	0.4028	0.3518	0.4773	0.6648	0.5813	0.7859
U-Net Ensemble	0.4926	0.3381	0.3791	0.3677	0.3913	0.6957	0.6746	0.7180
GB U-Net	0.5331	0.3909	0.4007	0.3509	0.4669	0.6862	0.6010	0.7997
(b) Nuclei segmentation results for the TNBC dataset
U-Net(VGG-16)	0.3538	0.1672	0.1614	0.1581	0.1648	0.5042	0.4940	0.5149
U-Net(VGG-19)	0.3829	0.1742	0.1364	0.1226	0.1536	0.5099	0.4585	0.5741
U-Net(ResNet-50)	0.3972	0.2324	0.2701	0.2530	0.2897	0.5638	0.5281	0.6048
U-Net(ResNet-101)	0.5080	0.3306	0.2904	0.2214	0.4217	0.5427	0.4139	0.7882
U-Net(ResNet-152)	0.4063	0.2594	0.2790	0.2478	0.3194	0.5874	0.5216	0.6723
U-Net(DenseNet-121)	0.4246	0.2929	0.3600	0.3327	0.3922	0.6216	0.5745	0.6772
U-Net(DenseNet-201)	0.4453	0.2843	0.3604	0.3287	0.3988	0.6298	0.5745	0.6970
U-Net(Inception-v3)	0.3817	0.1962	0.1562	0.1303	0.1947	0.4703	0.3925	0.5864
Mask R-CNN	0.4899	0.3449	0.4392	0.4027	0.4830	0.6732	0.6172	0.7403
U-Net Ensemble	0.4836	0.2808	0.2849	0.2430	0.3442	0.6068	0.5176	0.7331
GB U-Net	0.5403	0.3772	0.4205	0.3540	0.5176	0.6581	0.5541	0.8102

Bolded values identify the highest scoring segmntation method within 1)classical segmentation, 2)U-Net like DCNNs, 3)Mask R-CNN and ensemble networks, with respect to the evaluation metric.

Back to article page

Table 2 F1, recall, and precision metrics are reported for two intersection over union thresholds, 0.5 and 0.7.

Search

Quick links