Rekonstrukcija PET slika suparničkim modelima učenja

master's thesis

• Title (english): Reconstruction of PET Images Using Adversarial Learning Models
• Author: Marija Kuljanac
• Mentor: Damir Seršić (mentor)
• Committee member: Damir Seršić (predsjednik povjerenstva)
• Committee member: Marko Subašić (član povjerenstva)
• Committee member: Krešimir Križanović (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2022-07-12, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing

Abstract

PET predstavlja jednu od tehnika snimanja medicinskih slika te je u ovom diplomskom radu dan pregled teorijskih osnova nekih od najpoznatijih metoda njihove rekonstrukcije, FBP i MLEM. Zbog nedostataka upravo navedenih algoritama rezultati se pokušavaju poboljšati kombinacijom konvencionalnih metoda s metodama dubokog učenja, u ovom slučaju s generativnim suparničkim modelima. Dane su osnove strojnog učenja uz detaljniji pregled dvaju arhitektura generativnih neuronskih mreža - mreže za super rezoluciju (SRGAN) i rekonstrukciju (GANrec) kojima je zadatak poboljšati kvalitetu slika dobivenih na ulazu. Opisane su implementacije tih modela uz njihovu evaluaciju na parovima podataka koji se sastoje od slika iz online baze podataka i slika na kojima su simulirani nedostaci uređaja i klasičnih metoda rekonstrucije.

Abstract (english)

PET represents one of the medical imaging techniques and this master's thesis gives a theoretical overview of some of the most known methods of PET image reconstruction, FBP and MLEM. Because the methods mentioned have their faults there's an attempt to improve their results by combining them with deep learning, in this case with generative adversarial models. The basics of machine learning are given along with a detailed overview of two generative neural network architectures – super resolution generative network (SRGAN) and generative network for reconstruction (GANrec) whose main task is to improve the quality of input images. The implementations of these models are described as well as their evaluation on pairs of data made of images from the online database and simulated images that capture the imperfections of devices and ordinary reconstruction methods.

• Keywords: pozitronska emisijska tomografija; rekonstrukcija; duboko učenje; generativni suparnički modeli
• Keywords (english): positron emission tomography; reconstruction; deep learning; generative adversarial models
• Language: croatian
• URN:NBN: urn:nbn:hr:168:224000
• Study programme: Title: Information and Communication Technology; Study programme type: university; Study level: graduate; Academic / professional title: magistar/magistra inženjer/inženjerka informacijske i komunikacijske tehnologije (magistar/magistra inženjer/inženjerka informacijske i komunikacijske tehnologije)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Closed access
• Created on: 2023-01-14 09:14:27