Date of Award

Winter 11-2019

Document Type

Thesis

Degree Name

Bachelor Degree

Department

Mechatronics, System Engineering and Robotics

First Advisor

Bertan Karahoda

Language

Albanian

Abstract

By far, lung cancer is the prominent cause of cancer deaths for both men and women around the world. In 2018, statistics for WCRF (Worldwide Cancer Research Fund) showed that out of 2.09 million people diagnosed with this disease, 1.76 million people died. The survival rate increases if detected in its earlier stages. Taking into consideration the complexity of the problem, many computer-aided diagnosis systems that increase the survival rate have been proposed and developed. Driven by the notable success of deep learning in the area of complex image classification problems, this paper presents the use of VGG16, VGG19, ResNet34, and ResNet50 convolutional neural network architectures or classifying images of patients with cancer. Moreover, to compare the performance evaluation Accuracy, Precision, Area Under Curve, and F1 score were calculated. In conclusion, ResNet50 architecture exhibited the best result for this classification problem, with 95.83 Precision, 88.89% Accuracy, and 88.46% F1 score. The strategy of using pretrained deep learning models proved to be pertinent to this problem.

DOI

10.33107/ubt-etd.2019.190

Download

Share

COinS