Klasifikacija histopatoloških snimaka dijelova limfnih čvorova pomoću strojnog učenja

master's thesis

• Title (english): Classification of Histopathologic Scans of Lymph Node Sections Using Machine Learning
• Author: Domagoj Pluščec
• Mentor: Marko Čupić (mentor)
• Committee member: Marko Čupić (predsjednik povjerenstva)
• Committee member: Željka Mihajlović (član povjerenstva)
• Committee member: Alan Jović (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2019-07-10, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing

Abstract

Prilikom analize histopatoloških slika ukazuje se potreba za automatiziranim sustavom koji bi mogao pomoći doktorima u analizi slika i dijagnostici. Takav sustav bi mogao povećati točnost i brzinu analize i dijagnostike.
U okviru ovog rada dan je pregled područja analize histopatoloških slika te je izrađena programska implementacija za klasifikaciju limfnih čvorova temeljena na strojnom učenju.
Za učenje i testiranje modela strojnog učenja korišten je skup podataka PatchCamelyon.
Ispitani su rezultati koje postižu sljedeći modeli dubokog učenja: AlexNet, ResNet, DenseNet i Inception-v3 te kako na njihove rezultate utječu tehnike proširenja skupa podataka. Konačno najboljim se pokazao model Inception-v3 koji je na skupu za testiranje postigao točnost od 89%.

Abstract (english)

When analyzing histopathological images, there is a need for an automated system that could help doctors in image analysis and diagnostics. Such a system could increase the accuracy and speed of analysis and diagnostics.
Within this paper, an overview of the histopathological image analysis area is provided and a program implementation for the classification of lymph nodes based on machine learning has been developed.
PatchCamelyon dataset has been used for training and testing of chosen machine learning models.
The results of the following deep learning models have been studied: AlexNet, ResNet, DenseNet, Inception-v3. Also, the influence of different data augmentation methods on the model performance was investigated. Finally, the Inception-v3 model proved to be the best, which reached the 89% accuracy on the test set.

• Keywords: strojno učenje; analiza medicinskih slika; histopatologija
• Keywords (english): machine learning; medical image analysis; histopathology
• Language: croatian
• URN:NBN: urn:nbn:hr:168:732742
• Study programme: Title: Computing; Study programme type: university; Study level: graduate; Academic / professional title: magistar/magistra inženjer/inženjerka računarstva (magistar/magistra inženjer/inženjerka računarstva)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Closed access
• Created on: 2020-01-07 21:25:05