Register now After registration you will be able to apply for this opportunity online.
This opportunity is not published. No applications will be accepted.
Internship and/or master thesis: Developing Deep Learning Algorithms to Segment MR Images of the Knee
This project, collaboration between UZH and Balgrist Campus, aims to create a open-source tool for comprehensive segmentation of MR-knee. The end-goal of this project is to create a 'product' that will be used by the clinicians and researchers. If real-world impact is your thing, reach out to us. The attached project description (knee-seg-post.pdf) gives further details.
Keywords: deep-learning, magnetic resonance imaging, medical, image analysis, biomedical
Magnetic Resonance (MR) is one of the main in-vivo imaging techniques used to assess knee health. In the medical image analysis literature, there are several algorithms to segment knee tissues. However, each of them segment a certain subset of the components-of-interest. None of them simultaneously performs a comprehensive knee segmentation and is released as fully open and reproducible. So, the first objective of this project is to design algorithms that can learn from partially annotated datasets. Following this, we focus on deployment, focussing on performance and usability. Lastly, we aim to make the tool (partly) robust to resolution and contrast. This brings us into the realm of domain adaptation.
Magnetic Resonance (MR) is one of the main in-vivo imaging techniques used to assess knee health. In the medical image analysis literature, there are several algorithms to segment knee tissues. However, each of them segment a certain subset of the components-of-interest. None of them simultaneously performs a comprehensive knee segmentation and is released as fully open and reproducible. So, the first objective of this project is to design algorithms that can learn from partially annotated datasets. Following this, we focus on deployment, focussing on performance and usability. Lastly, we aim to make the tool (partly) robust to resolution and contrast. This brings us into the realm of domain adaptation.
To create an open-source, well-documented, and easy-to-use pipeline (e.g. a python library) to segment major knee tissues, including bone, cartilage, and meniscus from MR images. To this end, we aim to (1) implement the pipeline using existing partially segmented datasets to train and test the algorithms and leveraging on existing open-source code, and (2) adapt the segmentation algorithms to images of various resolutions.
To create an open-source, well-documented, and easy-to-use pipeline (e.g. a python library) to segment major knee tissues, including bone, cartilage, and meniscus from MR images. To this end, we aim to (1) implement the pipeline using existing partially segmented datasets to train and test the algorithms and leveraging on existing open-source code, and (2) adapt the segmentation algorithms to images of various resolutions.