TABLE 1.

Comparison between the proposed segmentation method and other single neural network.

	Diameter	U-net	nnUnet	U2Net	EG-net	Our method
Dice	50 > d ≥ 32 μm	0.680	0.710	0.753	0.758	0.801
	90 > d ≥ 50 μm	0.743	0.785	0.786	0.809	0.839
	d ≥ 90 μm	0.820	0.828	0.850	0.856	0.904
Jaccard	50 > d ≥ 32 μm	0.546	0.552	0.635	0.635	0.683
	90 > d ≥ 50 μm	0.619	0.624	0.683	0.701	0.737
	d ≥ 90 μm	0.707	0.718	0.763	0.767	0.832
Precision	50 > d ≥ 32 μm	0.722	0.681	0.777	0.781	0.824
	90 > d ≥ 50 μm	0.808	0.759	0.821	0.843	0.854
	d ≥ 90 μm	0.892	0.861	0.899	0.901	0.902
Sensitivity	50 > d ≥ 32 μm	0.700	0.715	0.779	0.780	0.802
	90 > d ≥ 50 μm	0.733	0.657	0.791	0.809	0.844
	d ≥ 90 μm	0.765	0.853	0.837	0.829	0.921