Fig. 2: Case study #1.
From: Deep-Manager: a versatile tool for optimal feature selection in live-cell imaging analysis
a A Standard feature extraction and discrimination capability evaluation using p-value for four handcrafted GFP intensity features {gmean, gtot, g25, and g75} computed over images acquired using fluorescence. n = 431 biologically independent samples have been considered. b Applying the DM platform to the four intensity features {gmean, gtot, g25, and g75} some descriptors reveal to be no more valid for the analysis, e,g,, g75, and g25 because too sensitive (SENS = 0.18 > thSENS (0.1) and SENS = 0.11 > thSENS (0.1) respectively) to perturbations (autofluorescence, photobleaching, saturation). Nevertheless, descriptor g75, still remains significant in terms of t-test analysis (p-value < ***) but presents a sensitivity value SENS larger than the threshold due to an unacceptable worsening in the DP performance after perturbation injection. n = 1293 biologically independent samples have been considered.
