Master thesis presentation: Segmentation of the Left Atrium using Watershed from Markers algorithm

Welcome to the following master thesis presentation:

Segmentation of the Left Atrium using Watershed from Markers algorithm
Presented by Reza Gharipour

13:00 at Wednesday September 4
Lunnerummet (room 3311), Hörsalsvägen 11, 3rd floor, Chalmers
Examiner: Yngve Hamnerius

Abstract
According to the statistics heart disease is the cause of more than 600,000 deaths per year in the USA. While 715,000 Americans annually have a heart stroke, it has been also the main cause of death in lots of other developing countries as well [1]. Among them, about 15% of all the heart attacks are associated with a condition called Atrial Fibrillation (AF). In case of permanent AF, Radio Frequency Catheter Ablation (RFCA) must be applied on the patient to prevent the risk of death. Hence knowing an exact geometry of the heart and its chambers is vital in success of the operation. Nevertheless, segmenting the Left Atrium and its pulmonary veins has been always a challenging task due to high variability of the Left Atrium. During the last years several methods have been proposed to overcome this situation.

Here we proposed a segmentation method based on Watershed from Markers algorithm, in order to extract a 3D geometry of the left atrium and its pulmonary veins. Also we compared the result of the method by its corresponding ground truth image (manual segmentation) to find the error rate of the method. The results implied that while the method is fast and parameter free, yet we encounter the oversegmentation issue in some cases. According to our observation, oversegmentation occurs mostly in the Left Ventricle (LV) and also the Pulmonary Veins (PVs). The overall result confirmed a mean overlap of 65-99% between the ground truth image and our method for all datasets.