Table 2.

Comparison of feature selection methods on field level features. The first two columns show rankings by different selection methods. RF returned 16 features while SDA with SVM classifier tuning returned 107 features.

RF	SDA	Feature name
1	1	Corr. between prot. and nuc. channels
2	2	Corr. between prot. and ER channels
3	64	Prot. and ER obj. overlap int. ratio
4	30	Threshold adjacency statistic #13
5	29	Prot. and nuc. obj. int. overlap ratio
6	11	Threshold adjacency statistic #12
7	7	Prot. and nuc. obj. area overlap ratio
8	10	Prot. and tub. obj. int. overlap ratio
9	4	Non-object fluorescence
10	3	Texture #16: corr. (4× downsampling)
11	37	Variance of # of pixels per object
12	35	Texture #3: corr. (4× downsampling)
13	5	Corr. between prot. and tub. channels
14	16	Info. between prot. and nuc. channels
15	12	Info, between prot. and ER channels
16	28	Ratio of largest to smallest obj. size