Register now After registration you will be able to apply for this opportunity online.
This opportunity is not published. No applications will be accepted.
Computer vision based reconstruction of neuromorphological features in the songbird’s syrinx on 3D tomograms (PSI)
Generation of speech in humans involves the modulation of the vocal cords in the larynx. Similarly, birds vocalize by modifying the membranes in the syrinx by muscular contraction. We aim to track syrinx´s muscle fibers and nerve cells through an ultra-high resolution 3D volume.
The generation of speech in humans involves the modulation of the vocal cords in the larynx. Similarly, birds vocalize by modifying the membranes in the syrinx by muscular contraction. While the general morphology of the syrinx is broadly understood (Düring et al., 2013), the neuromorphological fine structure is not known yet due to resolution limitation of non-invasive techniques.
We are studying how the syringeal muscles are innervated by performing 3D radiographic tomograms. This is a new cutting-edge imaging technique that uses synchrotron light (an intense X-ray beam) generated at the Swiss Light Source (a circular electron accelerator), part of the Paul Scherrer Institute at Villigen.
Available project
We offer a student project at the intersection of computer vision, machine learning, and neurobiology. In order to infer delicate neuromuscular control strategies of vocal production, we plan to track muscle fibers and nerve cells through an ultra-high resolution 3D volume represented by a stack of 2D images.
This project is pushing the boundaries of non-invasive, scientific, and biomedical imaging. To be able to distinguish nerve cells and muscle fibers from any background structure, a necessary first step is to segment the image. Your task would be to investigate how this can be done efficiently and with little to no ground truth data. You will deal with a highly challenging data set that needs to be analyzed exploratively. Possible approaches to segment the muscle fibers include edge detection combined with variations of watershed segmentation. To classify nerve cells, approaches could be to classify voxels using interactively trained random forests. Visual verification will be done to identify structures of interest by eye, in order to gauge the feasibility of the chosen approach. The aim of the project is to set up an image processing pipeline using machine learning and computer vision.
Background : You should have experience in computer vision and machine learning and preferably have worked on image segmentation before. Having used/implemented a watershed algorithm before is a distinct advantage. Students in computer sciences, physics or similar areas can apply. Knowledge in programming is a requirement.
The generation of speech in humans involves the modulation of the vocal cords in the larynx. Similarly, birds vocalize by modifying the membranes in the syrinx by muscular contraction. While the general morphology of the syrinx is broadly understood (Düring et al., 2013), the neuromorphological fine structure is not known yet due to resolution limitation of non-invasive techniques. We are studying how the syringeal muscles are innervated by performing 3D radiographic tomograms. This is a new cutting-edge imaging technique that uses synchrotron light (an intense X-ray beam) generated at the Swiss Light Source (a circular electron accelerator), part of the Paul Scherrer Institute at Villigen.
Available project
We offer a student project at the intersection of computer vision, machine learning, and neurobiology. In order to infer delicate neuromuscular control strategies of vocal production, we plan to track muscle fibers and nerve cells through an ultra-high resolution 3D volume represented by a stack of 2D images. This project is pushing the boundaries of non-invasive, scientific, and biomedical imaging. To be able to distinguish nerve cells and muscle fibers from any background structure, a necessary first step is to segment the image. Your task would be to investigate how this can be done efficiently and with little to no ground truth data. You will deal with a highly challenging data set that needs to be analyzed exploratively. Possible approaches to segment the muscle fibers include edge detection combined with variations of watershed segmentation. To classify nerve cells, approaches could be to classify voxels using interactively trained random forests. Visual verification will be done to identify structures of interest by eye, in order to gauge the feasibility of the chosen approach. The aim of the project is to set up an image processing pipeline using machine learning and computer vision.
Background : You should have experience in computer vision and machine learning and preferably have worked on image segmentation before. Having used/implemented a watershed algorithm before is a distinct advantage. Students in computer sciences, physics or similar areas can apply. Knowledge in programming is a requirement.
Not specified
Daniel Düring dnd@ini.ethz.ch
Nils Eckstein nilsec@ini.uzh.ch
Daniel Düring dnd@ini.ethz.ch Nils Eckstein nilsec@ini.uzh.ch