Finding Golgi Stacks in Electron Micrographs

Transmission electron microscopy (EM) can acquire images in which a range of subcellular organelles are clearly resolved simultaneously. There exist mature stereology techniques for extracting quantitative specimen information from section-based EM images and such techniques have been adopted successfully for use in immuno-EM. A bottleneck preventing the application of these nanomorphomics methods to high throughput applications is the recognition of organelle structures. This papers addresses this issue for one important organelle, the Golgi apparatus. A support vector machine is trained as a local Golgi detector based on rotationally invariant features. The SVM output is used to drive a graph-cuts segmentation. The ability of the method to detect and segment Golgi stacks is evaluated on a set of 36 micrographs.