Cell Detection Using Extremal Regions in a Semisupervised Learning Framework

Figure 2

Average F-score for varying size of training subimages for our method with only supervised loss (supervised), our semisupervised framework (SS), and our semisupervised framework using ILASTIK as the pixel classifier (SS + ILASTIK) is shown for all the testing datasets. The unlabeled data has contributions from both the training and testing images. We compare the performance of our method with that of Arteta et al. [1].

(a) Phase-contrast hela cells

(b) Synthetic fluorescence cells

(c) Drosophila Kc167 cells

(d) Fission yeast cells