Fig. 2: pySTED is used to artificially augment the training dataset of a DL model.
From: Development of AI-assisted microscopy frameworks through realistic simulation with pySTED
a, Segmentation task used in ref. 46 is used, in which the annotations comprise polygonal bounding boxes around F-actin fibres (magenta) and rings (green). b, pySTED is used to augment the training dataset by generating synthetic versions of a STED image. c, AP of the model for the segmentation of F-actin fibres (magenta) and rings (green). The model was trained on the original dataset from ref. 46 (O), and on the same dataset with updated normalization (N) and additional synthetic images (N + S). No significant changes in AP are measured for F-actin fibres, but a significant increase is measured for N + S over O and N for F-actin rings (Supplementary Fig. 6 shows the P values). d, Images were progressively removed from the dataset (100%, 42 images; 75%, 31 images; 50%, 21 images; 25%, 10 images; and 10%, 4 images). Removing more than 50% of the dataset for fibres negatively impacts the models, whereas removing 25% of the dataset negatively impacts the segmentation of rings (N; Supplementary Fig. 6 shows the P values). Adding synthetic images from pySTED during training allows 75% of the original training dataset to be removed without affecting the performance for both structures (N + S; Supplementary Fig. 6 shows the P values). Only the significant changes from the complete dataset are highlighted. The complete statistical analysis is provided in Supplementary Fig. 6. All the box plots show the distribution of five model training scenarios. The box extends from the first to the third quartile of the data, with a line at the median. The whiskers extend from the box to the farthest data point lying within 1.5× the interquartile range (IQR) from the box.
