Primerjava globokih modelov za segmentacijo membran v biomedicinskih tomografskih posnetkih elektronske mikroskopije

PREN, TIN (author), Bohak, Ciril (mentor), Žerovnik Mekuč, Manca (comentor)

V tem delu predstavimo rezultate poskusov, izvedenih z uporabo različnih orodij za segmentacijo mikroskopskih slik. V prvem delu predstavimo postopek tomografije, kako deluje, ter postopek izdelave tomografskih posnetkov. Podrobno predstavimo uporabljene testne podatke ter preizkušena orodja. Opišemo postopek testiranja in okolje, v katerem smo izvedli preizkuse, vključno s specifičnimi nastavitvami in prilagoditvami, potrebnimi za delovanje posameznih orodij. V drugem delu naloge predstavimo rezultate naših eksperimentov, podprte s slikovnim gradivom. Analiziramo učinkovitost in natančnost posameznega orodja ter diskutiramo prednosti in slabosti posameznih pristopov. Posebna pozornost je posvečena težavam, povezanim z uporabo zastarelih orodij, ki temeljijo na starejših knjižnicah in metodah, ter njihovim omejitvam.

• Language: Slovenian
• Keywords: segmentacija, tomografija, modeli globokega učenja
• Work type: Bachelor thesis/paper
• Typology: 2.11 - Undergraduate Thesis
• Organization: FRI - Faculty of Computer and Information Science
• Year: 2025
• PID: 20.500.12556/RUL-173273
• COBISS.SI-ID: 253632259
• Publication date in RUL: 15.09.2025

Secondary language

• Language: English
• Title: Comparison of deep models for membrane segmentation in biomedical tomographic electron microscopy images

Abstract

In this work, we present the results of experiments conducted using a variety of tools for the segmentation of microscopic images. In the first part, we describe the tomography process, explaining how it works and the steps necessary for the construction of tomographic images. We provide a detailed description of the test data and the tools used. The testing procedure and the environment in which the experiments were conducted are also described, including the specific settings and adjustments necessary for the functioning of each tool. In the second part of the work, we present the results of our experiments, supported by visual material. We analyze the effectiveness and accuracy of each tool and discuss the respective advantages and disadvantages of each approach. Special attention is given to the challenges associated with using dated tools that rely on older libraries and methods as well as their limitations.

• Keywords: segmentation, tomography, deep learning models