Automated cervical cell segmentation using deep ensemble learning

BMC Medical Imaging

Table 3 Performance comparison of base models, which encoders were initialized from ImageNet pre-trained models

Task Type	Model	Dice	Sensitivity	Specificity
Cytoplasm	Unet_resnet34	0.9497	0.9596	0.9819
	Unet_densenet121	0.9527	0.9625	0.9828
	UnetPlusPlus_resnet34	0.9533	0.9616	0.9835
	UnetPlusPlus_densenet121	0.9525	0.9594	0.9837
	DeepLabV3_resnet34	0.9407	0.9496	0.9795
	DeepLabV3_resnet50	0.9386	0.9492	0.9783
	DeepLabV3Plus_resnet34	0.9455	0.9475	0.9833
	DeepLabV3Plus_resnet50	0.9494	0.9598	0.9817
Nucleus	Unet_resnet34	0.7411	0.9431	0.9951
	Unet_densenet121	0.7506	0.9566	0.9952
	UnetPlusPlus_resnet34	0.8055	0.9481	0.9967
	UnetPlusPlus_densenet121	0.7731	0.9653	0.9957
	DeepLabV3_resnet34	0.6088	0.947	0.9906
	DeepLabV3_resnet50	0.6419	0.9506	0.9918
	DeepLabV3Plus_resnet34	0.6721	0.9053	0.9936
	DeepLabV3Plus_resnet50	0.7353	0.9483	0.9949

In the first column, cytoplasm and nucleus stand for the cytoplasm segmentation task and the nucleus segmentation task, respectively. Bold values represent the best results

ISSN: 1471-2342