Klasifikacija rendgenskih slika pluća primjenom konvolucijskih neuronskih mreža

undergraduate thesis

• Title (english): Classification of Chest X-Ray Images Using Convolutional Neural Networks
• Author: Petra Ilić
• Mentor: Zoran Kalafatić (mentor)
• Committee member: Zoran Kalafatić (predsjednik povjerenstva)
• Committee member: Petra Bevandić (član povjerenstva)
• Committee member: Matej Grcić (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2021-07-07, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing

Abstract

Pandemija koronavirusa izbila je krajem 2019. godine u Wuhanu u Kini i ubrzo postala ozbiljan problem u cijelom svijetu. Iako su razvijena cjepiva, još uvijek je ozbiljan problem i svakodnevno obolijeva znatan broj ljudi. PCR test i antigenski test najčešće su korištene metode detekcije COVID-a. Antigenski ili brzi test manje je osjetljiv i daje bolje rezultate kod osoba koje su već razvile simptome. PCR test je znatno osjetljiviji, ali često nije prvi izbor zbog visoke cijene i dužeg vremena čekanja. Rendgen je često korištena metoda detekcije raznih bolesti, pa se pojavilo pitanje može li se koristiti i za detekciju COVID-a. Pokazalo se da rendgenske slike pluća oboljelih od COVID-a 19 doista pokazuju specifične, ali teško uočljive abnormalnosti. Uvođenje automatske detekcije moglo bi znatno olakšati posao radiologa. Konvolucijske neuronske mreže pokazale su veliki potencijal u zadacima računalnog vida, pa tako i u klasifikaciji medicinskih slika. U ovom radu istražena je mogućnost primjene prijenosnog učenja za zadatak detekcije COVID-a 19. Četiri modela (ResNet50, ResNet101, AlexNet, DenseNet) korištena su kao ekstraktori značajki. Modeli su trenirani na skupu podataka COVID-XRay-5k i napravljena je evaluacija rezultata testiranja. Rezultati su ohrabrujući i pokazuju da bi konvolucijske neuronske mreže mogle biti od velike koristi u rješavanju ovog problema. Međutim, skup pozitivnih primjera je ograničen, pa se ne može tvrditi da su dobiveni rezultati pouzdani. Potrebno je napraviti detaljnije testiranje s većim i uravnoteženijim skupom podataka kako bi se dobili pouzdaniji podaci o preciznosti modela.

Abstract (english)

The coronavirus pandemic broke out at the end of 2019 in Wuhan, China, and soon became a serious problem worldwide. Although vaccines have been developed, it is still a serious problem and a significant number of people are getting infected daily. PCR test and rapid antigen test are the most used COVID-19 detection methods. Rapid antigen test is less sensitive and performs best in symptomatic people. PCR test is significantly more sensitive, but often it is not the first choice due to higher prices and longer waiting time. X-ray is one of the most commonly used illness detection methods, so naturally, the question of whether it can be used for COVID-19 detection arises. It has been shown that chest X-ray images of COVID-19 patients show specific, but hardly noticeable abnormalities. Automatic detection could be of great help to radiologists in diagnosing COVID-19 cases. Convolutional neural networks have shown great potential in solving computer vision problems, including the classification of medical images. In this study, transfer learning is proposed as a method for solving automatic COVID-19 detection problem. For this purpose, four popular models (ResNet50, ResNet101, AlexNet, DenseNet) have been used as feature extractors. They have been trained using the COVID-XRay-5k dataset and an evaluation of performance for each model has been made. The results are encouraging, as they show the promise of using CNN-s in COVID-19 detection. However, due to the limited number of positive examples, it cannot be claimed that the obtained results are reliable. More detailed testing using a larger and more balanced dataset is needed for a more reliable estimation of the accuracy of these models.

• Keywords: duboko učenje; prijenosno učenje; konvolucijske neuronske mreže; rendgen; koronavirus; automatska detekcija
• Keywords (english): deep learning; transfer learning; convolutional neural networks; x-ray; coronavirus; automatic detection
• Language: croatian
• URN:NBN: urn:nbn:hr:168:396767
• 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: 2023-01-14 20:37:34