Cell Detection Using Extremal Regions in a Semisupervised Learning Framework

Table 2

Quantitative comparison of detection for different datasets using our semisupervised approach using ILASTIK as a pixel classifier for varying number of subimages. Left-right: number of subimages, results using [1], SS + ILASTIK (unlabeled data has contributions from only the training images), and SS + ILASTIK (unlabeled data has contributions from both the training and testing images).

(a) Phase-contrast hela cells

	Arteta et al. [1]			SS + ILASTIK (training)			SS + ILASTIK (training + testing)
Number	Prec	Rec	F-score	Prec	Rec	F-score	Prec	Rec	F-score
1	0.9065	0.9253	0.9158 ± 0.0203	0.9180	0.9351	0.9264 ± 0.0082	0.9266	0.9498	0.9381 ± 0.0078
3	0.9342	0.9497	0.9419 ± 0.0084	0.9360	0.9511	0.9434 ± 0.0057	0.9397	0.9580	0.9488 ± 0.0064
5	0.9478	0.9545	0.9511 ± 0.0092	0.9480	0.9568	0.9523 ± 0.0061	0.9498	0.9599	0.9548 ± 0.0066
7	0.9481	0.9594	0.9537 ± 0.0091	0.9500	0.9596	0.9547 ± 0.0091	0.9513	0.9620	0.9566 ± 0.0076
9	0.9536	0.9603	0.9570 ± 0.0055	0.9528	0.9632	0.9579 ± 0.0080	0.9531	0.9661	0.9596 ± 0.0050

(b) Synthetic fluorescence cell images

	Arteta et al. [1]			SS + ILASTIK (training)			SS + ILASTIK (training + testing)
Number	Prec	Rec	F-score	Prec	Rec	F-score	Prec	Rec	F-score
1	0.9695	0.9527	0.9620 ± 0.0106	0.9730	0.9611	0.9670 ± 0.0034	0.9791	0.9698	0.9744 ± 0.0026
3	0.9716	0.9626	0.9671 ± 0.0086	0.9755	0.9662	0.9708 ± 0.0015	0.9801	0.9712	0.9756 ± 0.0005
5	0.9787	0.9643	0.9714 ± 0.0051	0.9794	0.9689	0.9741 ± 0.0013	0.9806	0.9723	0.9764 ± 0.0003
7	0.9792	0.9651	0.9721 ± 0.0036	0.9800	0.9699	0.9749 ± 0.0009	0.9811	0.9732	0.9771 ± 0.0006
9	0.9808	0.9652	0.9730 ± 0.0009	0.9806	0.9701	0.9753 ± 0.0010	0.9813	0.9750	0.9789 ± 0.0004

(c) Drosophila Kc167 cells

	Arteta et al. [1]			SS + ILASTIK (training)			SS + ILASTIK (training + testing)
Number	Prec	Rec	F-score	Prec	Rec	F-score	Prec	Rec	F-score
1	0.8195	0.9275	0.8702 ± 0.0156	0.8254	0.9306	0.8748 ± 0.0069	0.8311	0.9496	0.8864 ± 0.0043
3	0.8287	0.9337	0.8781 ± 0.0145	0.8358	0.9416	0.8855 ± 0.0082	0.8402	0.9566	0.8946 ± 0.0056
5	0.8402	0.9380	0.8864 ± 0.0083	0.8433	0.9452	0.8913 ± 0.0093	0.8451	0.9568	0.8975 ± 0.0080
7	0.8427	0.9435	0.8903 ± 0.0055	0.8463	0.9475	0.8940 ± 0.0070	0.8499	0.9580	0.9007 ± 0.0032
9	0.8516	0.9456	0.8962 ± 0.0045	0.8519	0.9497	0.8981 ± 0.0052	0.8526	0.9590	0.9028 ± 0.0038

(d) Fission yeast cells

	Arteta et al. [1]			SS + ILASTIK (training)			SS + ILASTIK (training + testing)
Number	Prec	Rec	F-score	Prec	Rec	F-score	Prec	Rec	F-score
1	0.7492	0.9238	0.8274 ± 0.0136	0.8064	0.9230	0.8607 ± 0.0074	0.8528	0.9243	0.8871 ± 0.0053
3	0.7602	0.9342	0.8383 ± 0.0112	0.8145	0.9276	0.8673 ± 0.0089	0.8602	0.9245	0.8912 ± 0.0050
5	0.7622	0.9412	0.8423 ± 0.0076	0.8226	0.9414	0.8779 ± 0.0076	0.8707	0.9409	0.9044 ± 0.0051
7	0.7645	0.9494	0.8470 ± 0.0039	0.8258	0.9465	0.8820 ± 0.0062	0.8736	0.9420	0.9065 ± 0.0039
9	0.7718	0.9460	0.8501 ± 0.0046	0.8287	0.9460	0.8834 ± 0.0058	0.8791	0.9470	0.9118 ± 0.0047