Fig. 6: Enhanced membrane segmentation using MemBrain v2. | Communications Biology

Fig. 6: Enhanced membrane segmentation using MemBrain v2.

From: cryoTIGER: deep-learning based tilt interpolation generator for enhanced reconstruction in cryo electron tomography

Fig. 6

A Representative tomogram slice from a tilt series downloaded from the cryo-ET Data Portal47 and reconstructed using our pipeline (left). The same slice with the hybrid ground truth segmentation (manually curated) is depicted in white (middle). The tomogram slice overlaid with the GT segmentation (right). B Comparison of the non- interpolated version with interpolation using the DL (Vimeo) model and the DL (cryo-ET) model. Green regions represent areas in the GT but not detected by MemBrain v2 (false negatives), red regions were detected by MemBrain v2 but are not in the GT (false positives), and blue regions were detected by both MemBrain v2 and the GT (true positives). C Overlap-based dice similarity coefficient for four tested tomograms, ranging from 0 (no overlap) to 1 (full overlap). D Overlap-based Jaccard Index for the same tomograms, also ranging from 0 to 1. E Hausdorff distance measuring how far the tested segmentation outlines are from the GT segmentation. Scale bar: 60 nm.

Back to article page