Figure 6
From: Label-free imaging flow cytometry for analysis and sorting of enzymatically dissociated tissues
MLP training and assessment. (A) Plot shows a learning rate screening for all MLP architectures. During screening, MLPs are trained using the available training data and data augmentation methods are applied. The learning rate screening was performed using AIDeveloper 0.2.3. (B) Plot shows the validation accuracy during training of four MLPs to distinguish GFP- and GFP+ cells. For a smooth appearance, each line shows the rolling median (window size = 50). (C) Green and gray histogram show the probabilities returned by MLP2 for each event the GFP+ and GFP- class of the validation set. (D) Scatterplot shows the concentration and yield of GFP+ rod photoreceptors when applying MLP2 to the validation set using different threshold values P(GFP+)thresh for prediction. (E) Confusion matrices when using a threshold P(GFP+)thresh of 0.5 and 0.67. The red rectangle indicates the events that are predicted to be GFP+. Those events would be sorted during a sorting experiment, resulting in a particular concentration of GFP+ cells (cGFP+) in the sorted sample.
