Table 4.

Comparison of the performance of the different channel fusion methods on the test set images, assessed with F1-score as a segmentation metric.

Organelle	Fusion method	CL1	CL2	CL3	CL4	CL5
		0, 1, 2 (CW)	0, 1, 2, 3 (CW)	0, 1, 2, 3 (CW)	0, 1, 2, 3 (CW)	0, 1, 2 (CW)
Cytoplasm	FA	0.8135	0.5138	0.3264	0.7769	0.6873
	SIMPLE	0.729	0.4744	0.2967	0.6846	0.4218
	STAPLE	0.8043	0.4043	0.2876	0.6881	0.6635
	Voting	0.729	0.4571	0.3094	0.6916	0.4218
	Majority Voting	0.729	0.4344	0.3122	0.7035	0.4218
Nuclei	FA	0.904	0.6871	0.843	0.8758	0.7994
	SIMPLE	0.8016	0.3804	0.7634	0.7303	0.638
	STAPLE	0.8016	0.4398	0.7341	0.6822	0.6449
	Voting	0.8016	0.2968	0.7194	0.6784	0.6449
	Majority Voting	0.8016	0.3266	0.6941	0.6817	0.6449

Note. It must be noted that the comparison is drawn between the fusion of all channels for each cell lines, evaluated using the Channel-wise strategy—and not the best scoring strategy or set of training/evaluation channels. Although the fusion method performance holds for other variants of our overall method.