Fall Semester:I am teaching the image analysis and computer vision course together with Prof. Van Gool, Prof. Szekely and Prof. Goksel.
Sprint Semester: Starting Sprint 2018, I will be teaching Medical Image Analysis together with Prof. Philippe Cattin and Prof. Mauricio Reyes.
Below there is a short list of currently available projects in our lab. Please contact me about more information on these projects as well as other available ones, which are not listed here.
Optical coherence tomography (OCT) is a non-invasive, in vivo imaging tool that uses near-infrared light to allow reliable, high resolution quantification of retinal layer thicknesses. It is a tool commonly used in the clinical setting for diagnosis and for monitoring disease progression or treatment response in ophthalmological conditions (ie. glaucoma) and even in neurological disorders such as multiple sclerosis.
Recently, the tool has been modified for pre-clinical research in animal models to further examine the biological mechanisms involved in retinal degeneration. As research in this area is relatively new, currently available tools for segmenting and identifying the different retinal layers from the OCT measurements has not yet been fully optimized for rodent imaging. Although manual segmentation has been found to be reproducible, it is extremely time consuming and prone to biases. Software for human can be applied to rodent images, however, as the image characteristics change the algorithm accuracy decreases and resulting layers require extensive manual interaction to correct. Therefore, there is a need for an automated OCT segmentation tool that can accurately and precisely assess rodent data.
The goal of this project is to develop machine learning based algorithms for extracting the layers from the OCT data acquired from small animals. The student is expected to investigate various approaches based on convolutional neural networks and recurrent neural networks to address the problem. Interest in machine learning based image analysis is essential. The student is expected to have some experience with Python programming language and experience with deep learning model is a plus.
The project is a collaboration with the group of Prof. Sven Schippling at UZH and Dr. Praveena Manogaran. The tools developed within the framework of this project will be directly used in Dr. Manogaran's research to advance our understanding in Multiple Sclerosis. The level of the proposed project is at the Master's level. However, motivated semester project students are encouraged to apply.
Calcific Aortic Stenosis (AS) is a condition in which calcium build-up along the aortic valve leads to severe obstruction. It is one of the most common heart valve disorders in developed countries. The condition is serious as it can restrict blood flow through the body and potentially even cause heart failure. For patients with sever AS, Surgical Aortic Valve Replacement (SAVR) is a common treatment
Transcatheter Aortic Valve Replacement (TAVI) surgery is a less invasive alternative to SAVR. Artificial valve is inserted through the femoral artery and guided to the aortic valve position without open heart surgery. Due to its minimally invasive nature, TAVI has benefits over traditional surgery. However, as compared to traditional open-heart valve replacement surgery, TAVI currently poses the risk of post-surgery complications, such as aortic regurgitation, in which the new valve slips from its inserted position. As these complications may be life-threatening, there is a need to better understand their causes and predict the likelihood of their occurrence.
Calcification around the aortical valve is an important factor for determining risk factors for individuals undergoing TAVI surgery. In this project, the goal is to build machine learning based algorithms to automatically detect calcification around the aortical valve. Detections will then be used to predict surgical outcomes with the hope of identifying risky individuals before the surgery.
Interest in machine learning based image analysis is essential. Experience with Python programming language and deep learning methods is a plus but not necessary. This is a continuation of a previous MS project. This project is a collaboration between ETHZ and USZ.
The level of the proposed project is at the MS level. However, motivated semester project students are encouraged to apply.
Generative adversarial networks (GAN) are a recent development which have shown great success in improving the performance of deep learning models for image classification. In short, GAN can generate artificial images that look like natural or actual images to the human observer (or the contesting network). The purpose of this project is to implement a GAN to inject or remove certain features suggestive for a disease into given medical images, good enough to fool trained medical doctors.
This is a joint project between the Computer Vision Laboratory, ETH and Dr. Anton Becker from the Institute of Diagnostic and Interventional Radiology, USZ.