A large part of the medical imaging data available today is available in 3D. This is true for example for magnetic resonance imaging data, computed tomography scan data and many other techniques. This is advantageous as it allows drawing conclusions based not only on single images but taking into account more information. Many medical images have to be further processed by segmenting different objects in the images such as bones or tissue layers. In the past years, the analysis of 2D medical images with Deep Learning methods has become a standard in research. With more and more computing power available, the focus is moving towards 3D analysis to profit from more spatial context.
The student will work with stacks of optical coherence tomography (OCT) images of the eye. These images are segmented into different layers and can be used to detect diseases and abnormalities or simply evaluate changes over time. In our projects, these images are segmented into several layers with the help of Convolutional Neural Networks (CNNs).The goal of this project is to determine the advantages and disadvantages of 3D CNNs for this type of images. Therefor the student will experiment with different 3D Networks and assess their results for several different types of segmentations. We will use segmentations generated from an existing 2D Network as ground truth for training and evaluation as well as the output of current state-of-the-art OCT machines.
- Fluency in a high level programming language (e.g. Python)
- Computer Vision basics very desirable
- Machine/Deep Learning knowledge helpful, willingness to learn about frameworks
Kind of Work
30% Research, 40% Software Development, 30% Benchmarking
Python/Matlab, Machine Learning (desirable), Deep Learning Frameworks (e.g. TensorFlow, mxNet)
Time & Effort
Semester Thesis, 1 Person