Densely Populated Cell and Organelles Segmentation with U-Net Ensembles
Fulton, S.; Baenen, C.; Storrie, B.; Yang, Z.; Leapman, R. D.; Aronova, M.
Show abstract
The complex and highly intertwined morphology of activated platelets within thrombi poses significant challenges for segmentation. In this work, we present a robust dual-network pipeline for cell and organelle segmentation. This multi-network approach enables the detection of fine details near the membrane while simultaneously facilitating long-range smoothing in regions distal to the membrane, drastically improving the performance of the watershed clustering algorithm compared to single-network approaches. We further enhance segmentation performance by collecting neural network predictions along multiple axes, capturing 3D correlations using only 2D neural networks. We segment and analyze hundreds of platelets and report quantitative morphological measurements, showing volumes consistent with hand-segmented results. We apply our segmentation pipeline to the CREMI neuron segmentation challenge data and provide state-of-the-art results.
Matching journals
The top 13 journals account for 50% of the predicted probability mass.
Similar papers in this journal
- DeLTA: Automated cell segmentation, tracking, and lineage reconstruction using deep learning 96%
- DeepMIB: User-friendly and open-source software for training of deep learning network for biological image segmentation 95%
- VolPy: automated and scalable analysis pipelines for voltage imaging datasets 95%
Similar papers in this journal
- Towards unsupervised classification of macromolecular complexes in cryo electron tomography: challenges and opportunities 94%
- MemBrain: A Deep Learning-aided Pipeline for Automated Detection of Membrane Proteins in Cryo-electron Tomograms 93%
- A deep learning workflow for quantification of Micronuclei in DNA damage studies in cultured cancer cell lines: a proof of principle investigation 91%
Similar papers in this journal
Similar papers in this journal
Similar papers in this journal
"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.