Master thesis

My Master of Engineering master thesis "Liver Perfusion using Level Set Methods" has been written in between September 2004 and May 2005 at the Shanghai Jiaotong University under the supervision of Professor Lixu Gu at the Medical Image Processing and Visualization laboratories.

The thesis was graded excellent.

Abstract

The clinical process known as liver perfusion is a well-established method to determine blood flow conditions. For certain illnesses related to the liver, information about the blood flow to the liver can help in obtaining an accurate diagnosis. The process is then called liver perfusion and works as follows. A series of MRI images of the patients abdomen is obtained in regular time intervals. At some point, a contrast agent is injected into the patients blood circuit, creating a visible response in the MRI. By tracking the response at the liver, the perfusion intensity curve over time can be deduced.

This thesis describes a novel method of automatically obtaining the perfusion intensity curve from a series of MRI images. After a minimal manual initialization, the system automatically segments and marks the liver in the image and tracks the liver over time across all images. For the segmentation - the marking of the liver shape inside the image - both the modern Fast Marching Method and the Level Set Method are used. By using a flexible framework combining both these methods and by choosing a segmentation speed function carefully, a good performance is achieved in both accuracy and runtime performance. A fully functional prototype application demonstrating the liver perfusion measurement using minimal manual initialization has been developed and tested.

The results are validated on more than ten real clinical series, by feedback from radiologists and by two synthetic tests. In all evaluations the method proves to be significantly better than the previously used method. The improvement stems from the more careful model being used to register images in the series; moreover efficient interpolation methods are used to compensate for outliers, producing a good perfusion curve.

Download

thesis-nowozin-2005.pdf (13.6 Mb)


last update: Sun, 2nd July 2006