diplomsko delo
Blaž Palčnik (Author), Nejc Mekiš (Reviewer), Janez Žibert (Mentor)

Abstract

Uvod: Sistemi za računalniško podprto diagnozo (CAD) so se razvili za pomoč zdravnikom radiologom pri interpretaciji računalniško-tomografskih (CT) slik. V tem diplomskem delu sta opisana CAD sistem za klasifikacijo pljučnih vozličkov in CAD sistem za klasifikacijo jetrnih patologij. Sistemi so nastali s pomočjo strojnega učenja. Običajno so sestavljeni iz štirih stopenj. Prva stopnja je postopek predhodne obdelave, ki se izvaja z namenom izboljšanja kakovosti CT slik. Naslednji korak je postopek segmentacije, ki je pomemben za izločitev pomembnih lastnosti s CT slik. Ta korak nam omogoča, da smo lahko na sliki pozorni le na njene pomembne dele. Postopek klasifikacije pomeni združevanje slik z enakimi lastnostmi v razrede. Je zelo kompleksen proces, za katerega moramo zagotoviti ogromno količino podatkov za učenje algoritmov. To dosežemo s podatkovnimi zbirkami, ki so lahko prosto dostopne ali ne. Ločiti jih moramo na učne in testne podatke. Uspešnost CAD sistemov merimo s kontingentično tabelo, iz katere izračunamo specifičnost, senzitivnost in točnost. Zaželena je tudi ROC krivulja in izračun območja pod krivuljo (AUC). Namen: Namen diplomskega dela je bil pregled strokovne literature, ki se navezuje na uporabo CAD aplikacij v računalniški tomografiji pri slikanju prsnih organov in jeter. Glavni cilj je bil s pomočjo sistematičnega pregleda literature poiskati uporabo, delovanje, uspešnost in ključne sestavne dele CAD sistemov. Metode dela: Uporabljena je bila deskriptivna metoda dela, s sistematičnim pregledom literature na področju medicine in računalništva. V rezultate smo vključili 5 člankov, ki so se nam zdeli najbolj ustrezni. Rezultati: V rezultatih je predstavljeno: 2 CAD sistema za odkrivanje pljučnih vozličkov in 3 sistemi za odkrivanje različnih patologij na jetrih. Sistemi so bili testirani v različnih podatkovnih bazah. Opis in uspešnost sistemov je predstavljena v obliki tabele. Razprava in zaključek: CAD sistemi dosegajo glede na pregledano literaturo izjemne rezultate. Imajo pa tudi določene slabosti. Sklepamo, da najboljše rezultate dosegajo sistemi, ki uporabljajo algoritme globokega učenja. Omejitve pri primerjanju rezultatov predstavljajo raznolikost CT slik in podatkovnih baz. Večje implementacije CAD sistemov v klinični praksi še nismo zasledili. Za to bo potrebno še nekaj časa, vendar verjamemo, da se bodo uporabljali.

Keywords

diplomska dela;radiološka tehnologija;računalniška tomografija;strojno učenje;CAD sistemi;pljučni vozlički;jetra;

Data

Language: Slovenian
Year of publishing:
Typology: 2.11 - Undergraduate Thesis
Organization: UL ZF - University College of Health Studies
Publisher: [B. Palčnik]
UDC: 616-07
COBISS: 76160003 Link will open in a new window
Views: 246
Downloads: 51
Average score: 0 (0 votes)
Metadata: JSON JSON-RDF JSON-LD TURTLE N-TRIPLES XML RDFA MICRODATA DC-XML DC-RDF RDF

Other data

Secondary language: English
Secondary title: An overview of the use of CAD applications in computed tomography
Secondary abstract: Introduction: Computer-aided diagnosis (CAD) systems have been developed with purpose to help doctors, primarily radiologists, in computer tomography (CT) image interpretation. In this diploma work, CAD systems for lung nodule classification and classification of liver pathologies are described. CAD systems were developed using machine learning. It usually contains four stages. The first stage is image preprocessing, to enhance image quality. The next stage is segmentation, to remove interesting objects from other image data. In the classification process, images with the same properties are combined into classes. It is a very complex process, for which we need to provide a huge amount of data to learn the relevant algorithms. We can provide it, using public or private databases. We need to seperate the information into training and testing data. We can calculate performance of systems using a confusion matrix. We can calculate sensitivity, specificity and accuracy. We can also measure performance using receiver operating characteristics curve (ROC) and value of area under the curve (AUC). Purpose: In this diploma work, systematic overview has been performed for CAD applications in CT lung and liver imaging. The main purpose was to review the architecture of CAD systems, its implementation in clinical practice and evaluate its performance Methods: In diploma work, we used a descriptive method and systematic analysis of numerous scientific articles from computer science and medicine. The five most relevant articles are included in the results. Results: Two CAD systems for lung nodule classification and three systems for the diagnosis of various liver pathologies are presented and systems have been tested on different databases. The description and evaluation of the system are tabulated. Discussion and conclusion: CAD systems are achieving great results. However, we have to bear in mind that there are some limitations. We conclude that the best results are achieved by systems that use deep learning algorithms. There are also problems in comparing results of CAD systems, because of the diversity of CT images and databases. A huge number of potentional application of CAD systems in clinical practice have not been reported yet, but we believe they will be applied in the future.
Secondary keywords: diploma theses;radiologic technology;computed tomography;machine learning;CAD systems;lung nodules;liver;
Type (COBISS): Bachelor thesis/paper
Study programme: 0
Thesis comment: Univ. v Ljubljani, Zdravstvena fak., Oddelek za radiološko tehnologijo
Pages: 38 str.
ID: 13386203