Klasifikacija histoloških snimaka crijeva predtreniranim neuronskim mrežama

undergraduate thesis

• Title (english): Classification of histological images of the intestine using pretrained neural network
• Author: Dominik Poljak
• Mentor: Marko Subašić (mentor)
• Committee member: Marko Subašić (predsjednik povjerenstva)
• Committee member: Donik Vršnak (član povjerenstva)
• Committee member: Ivana Matovinović (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2024-07-01, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing

Abstract

U ovom radu istražujemo primjenu konvolucijskih neuronskih mreža i prijenosnog učenja na zadatku klasifikacije Crohnove bolesti i ulceroznog kolitisa na temelju histoloških snimaka. Fokus je na treniranju predtrenirane mreže EfficientNet, te obradi i prilagodbi ulaznih podataka i njihovoj podjeli na skupove za treniranje, validaciju i testiranje. Rad daje pregled i objašnjenje neuronskih mreža i teorije prijenosnog učenja. Prolazimo kroz postupke stvaranja zakrpa, multi-Otsu automatske segmentacije, otkrivanja bitnih dijelova slike, čišćenja nevažećih podataka i normalizacije. Prikazani su ključni hiperparametri koji utječu na performanse modela te njihov odabir za ovu vrstu problema. Rezultati analize prikazuju učinkovitost različitih modela s različitim konfiguracijama u razlikovanju između dvije navedene bolesti.

Abstract (english)

In this paper, we investigate the application of convolutional neural networks and transfer learning to the task of classifying Crohn's disease and ulcerative colitis based on histological images. The focus is on training a pretrained EfficientNet model, as well as processing and adapting the input data and dividing it into training, validation, and testing sets. The paper provides an overview and explanation of neural networks and the theory of transfer learning. We go through the procedures of patch creation, multi-Otsu automatic segmentation, identification of important image regions, cleaning invalid data, and normalization. Key hyperparameters that affect model performance and their selection for this type of problem are presented. The results of the analysis show the effectiveness of different models with various configurations in distinguishing between the two mentioned diseases.

• Keywords: računalni vid; duboko učenje; konvolucijske neuronske mreže; prijenosno učenje; EfficientNet; binarna klasifikacija; automatska segmentacija slike; histološke slike; H&E normalizacija
• Keywords (english): computer vision; deep learning; convolutional neural networks; transfer learning; EfficientNet; binary classification; automatic image segmentation; histological images; H&E normalization
• Language: croatian
• URN:NBN: urn:nbn:hr:168:545943
• 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: 2024-10-12 21:53:44