Semantička segmentacija slika magnetske rezonance tumora na mozgu pomoću strojnog učenja

undergraduate thesis

• Title (english): Semantic Segmentation of Magnetic Resonance Images of Brain Tumors Using Machine Learning
• Author: Fran Žužić
• Mentor: Marin Šilić (mentor)
• Committee member: Marin Šilić (predsjednik povjerenstva)
• Committee member: Helena Čeović (član povjerenstva)
• Committee member: Zorana Doždor (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2023-06-28, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing

Abstract

U sklopu ovog rada razmotren je problem segmentacije medicinskih slika magnetske rezonance tumora na mozgu. Razmotrena je dvodimenzionalna i trodimenzionalna segmentacija. Primijenjen je relevantan pristup segmentiranja pomoću konvolucijskih neuronskih mreža, temeljenih na U-Net arhitekturi. Eksperimenti za navedene vrste segmentiranja medicinskih slika napravljeni su na dva različita skupa podataka. Rezultati dobiveni izradom i primjenom modela neuronskih mreža tabelirani su i grafički prikazani, uz prikaze segmentiranih slika. Raspravljeni su i problemi koji su česti u području segmentacije medicinskih slika kao što su veliki zahtjevi za memorijskim prostorom, neravnoteža klasa u ground truthu te mali broj označenih slika.

Abstract (english)

As part of this work, the problem of segmentation of medical magnetic resonance images of brain tumors was considered. Two-dimensional and three-dimensional segmentation were considered. A relevant segmentation approach using convolutional neural networks, based on the U-Net architecture, was applied. Experiments for the mentioned types of segmentation of medical images were made on two different data sets. The results obtained from the creation and application of neural network models are tabulated and graphically presented, along with segmented images. Problems that are common in the field of medical image segmentation, such as large memory space requirements, class imbalance in the ground truth and a small number of labeled images, are also discussed.

• Keywords: konvolucija; konvolucijske neuronske mreže; U-Net; magnetska rezonanca mozga; diceov koeficijent; segmentacija medicinskih slika; treniranje neuronske mreže
• Keywords (english): convolution; convolutional neural networks; U-Net; brain magnetic resonance imaging; dice coefficient; segmentation of medical images; neural network training
• Language: croatian
• URN:NBN: urn:nbn:hr:168:936047
• Study programme: Title: Computing; Study programme type: university; Study level: undergraduate; Academic / professional title: sveučilišni/a prvostupnik/ prvostupnica (baccalaureus/baccalaurea) inženjer/inženjerka računarstva (sveučilišni/a prvostupnik/ prvostupnica (baccalaureus/baccalaurea) inženjer/inženjerka računarstva)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Closed access
• Created on: 2023-10-11 11:06:09