Desarrollo de algoritmos de procesamiento de imagen avanzado para interpretación de imágenes médicas. Aplicación a segmentación de hígado sobre imágenes de Resonancia Magnética multisecuencia

Abstract

Medical imaging has become in recent years a powerful tool to support diagnosis. Thanks to advanced scanners and image reconstruction software available, it is possible to identify different organs and tissues, as well as obtaining data that may help to characterize and quantify the pathologies. Radiologists are responsible for the use and interpretation of these images and require tools that allow them to locate organs and tissues with greater accuracy and speed, as well as the identification and quantitative characterization of the pathologies present in them, in order to make an accurate diagnosis. Moreover, liver cancer is one of the leading causes of cancer death worldwide. Invasive techniques used for diagnosis, such as surgical biopsies can sometimes be replaced by noninvasive techniques in medical imaging such as computed axial tomography (CT) and magnetic resonance imaging (MRI), with clear benefits for the patient. In order to assist radiologists and surgeons in a reliable intervention planning, new methods and accurate and efficient tools are needed to locate and segment properly the organ of interest and the pathologies inside. Automatic segmentation (delimitation) of the liver is a complex problem. Partial results have been achieved mainly on images obtained by CT. MRI technique provides more information for diagnostic purposes. However, liver segmentation in MRI images is a challenge due to the presence of characteristic artifacts, such as the partial volumes, the noise, and generally, the low sharpness and the low contrast between organs, so that the boundary between different tissues is often confusing. There are fewer developments on MRI, although these have been steadily increasing in recent years. In this thesis, we present a novel method for multichannel MRI automatic liver segmentation. The proposed method consists of the minimization of a 3D active surface by means of the dual approach to the variational formulation of the underlying problem. This active surface evolves over a probability map that is based on a new compact descriptor comprising spatial and multisequence information of every pixel which is further modeled by means of a liver multivariate statistical model that has been previously generated. This proposed 3D active surface approach naturally integrates volumetric regularization in the statistical model. The advantages of the compact visual descriptor together with the proposed approach result in a fast and accurate 3D segmentation method. The method was tested on 18 healthy liver studies and results were compared to a gold standard made by expert radiologists. Comparisons with other state-of-the-art approaches are provided by means of nine well established quality metrics. The obtained results are in line with the state-of-the-art methodologies, and are even better than them in some cases. A Dice Similarity Coefficient of 98.59 has been achieved.

Date of Award	2016
Original language	English
Awarding Institution	Universidad del País Vasco (UPV/EHU)
Supervisor	Pedro Mª Iriondo Bengoa (Supervisor) & Artzai Picon Ruiz (Supervisor)