Supervised view type classification and left ventricle segmentation in 2D Cardiac Ultrasound images
Independent thesis Advanced level (degree of Master (Two Years)), 20 HE creditsStudent thesis
Heart ultrasound examination is the most common technology usedwhen it comes to assessing the cardiac function. The main indicators of this function are the shape and deformation of the left ventricle. In 2D ultrasounds, these indicators are mainly extracted from apical images, which are views of the heart where the left ventricle is most clearly visible and its function easier to estimate. There are three main apical views that can be observed depending on the position of the probe. Currently, all measurements extracted from these views require a manual interaction of the operator in order to segment the left ventricle or select which of the apical views is being observed, which has a significant impact on both the speed of the exam and the repeatability of the results among different operators. The aim of this thesis is to speed up this procedure and reduce the inter-operator variability, by making the image annotation task fully automatic. In this thesis, we will focus on two main tasks part of this global automation of heart function’s assessment: classification of the three apical views and segmentation ofthe left ventricle. We will use phase based descriptors and keypoints’ extractors in order to achieve both of these tasks.
Place, publisher, year, edition, pages
2016. , 38 p.
medical imaging, ultrasounds, heart, ventricle, segmentation, classification, deep learning, neural networks
IdentifiersURN: urn:nbn:se:kth:diva-184218OAI: oai:DiVA.org:kth-184218DiVA: diva2:915609
Subject / course
Master of Science - Machine Learning
2016-02-08, 14:11 (English)
Kragic, Danica, Professor