Fig. 2From: Application of convolutional neural networks towards nuclei segmentation in localization-based super-resolution fluorescence microscopy imagesSegmentation results using a Kaggle dataset trained Mask R-CNN applied to our STORM images. Segmented super-resolution images from the colon tissue dataset were originally downsized to 512 × 512 before testing (a), and then further downsized to 256 × 256 and blurred before segmenting again (b). The Kaggle trained network attained superior results on the blurred 256 × 256 sized image, however still demonstrated significant over segmentation. This result demonstrates the need to train these CNN segmentation methods on the super-resolution data directlyBack to article page