Description »

Electron Microscopy (EM) is a recent imaging technique that can produse very high resolution images that are critical to unlocking new insights into brain functions. Our lab recently focused on developing automatic 3D segmentation techniques for such large datasets. Having automatic algorithms is highly desirable as they allow biologists to avoid hours of painful manual annotations but interactivity is still required as fully automatic algorithms can not yet achieve the same accuracy as human experts. This project will consist of 3 steps:

1. 1. Experiment with different online machine learning algorithms like Random forest or svmsgd (Stochastic Gradient Descent for Support Vector Machines);
2. 2. Extend an existing QT interface for labelling different objects in the images. The labels will then be used to train one of the online machine learning algorithm chosen in Step 1;
3. 3. Optional: Once an object has been segmented, a 3d mesh can be constructed in order to extract statistics such as surface or volume areas.