Fast and Accurate Segmentation of Filaments and Membranes in Micrographs with TARDIS
Author Information
Author(s): Kiewisz Robert, Fabig Gunar, Conway Will, Johnston Jake, Kostyuchenko Victor A., Bařinka Cyril, Clarke Oliver, Magaj Magdalena, Yazdkhasti Hossein, Vallese Francesca, Lok Shee-Mei, Redemann Stefanie, Müller-Reichert Thomas, Bepler Tristan
Primary Institution: Simons Machine Learning Center, New York Structural Biology Center, New York, United States
Hypothesis
Can TARDIS provide a fast and accurate method for annotating filaments and membranes in electron microscopy images?
Conclusion
TARDIS significantly improves the speed and accuracy of segmenting biomolecular structures in electron microscopy data.
Supporting Evidence
- TARDIS improves annotation accuracy by 42% for microtubules and 55% for membranes compared to existing tools.
- TARDIS can annotate a single tomogram in minutes, significantly faster than manual methods.
- TARDIS has been applied to segment over 13,000 tomograms, a task that would take a human approximately 35 years to complete manually.
- TARDIS is open-source and freely available for the scientific community.
Takeaway
TARDIS is a smart tool that helps scientists quickly and accurately label tiny structures in pictures of cells, making their work much easier.
Methodology
TARDIS uses a machine-learning framework combining deep learning for semantic segmentation and a geometric model for instance segmentation.
Limitations
The study relies on a manually annotated dataset, which may not cover all possible structures.
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website