Unit:
Department of Informatics and TelecommunicationsΠληροφορική
Author:
Bartsokas Theodoros
Supervisors info:
Παναγιώτης Σταματόπουλος, Επίκουρος Καθηγητής, Τμήμα Πληροφορικής και Τηλεπικοινωνιών, Εθνικό και Καποδιστριακό Πανεπιστήμιο Αθηνών
Original Title:
Κατάτμηση και Kατηγοριοποίηση Eγκεφαλικών Όγκων με χρήση Βαθιάς Μάθησης
Translated title:
Brain Tumor Segmentation and Classification using Deep Learning
Summary:
The purpose of this paper is the development of methods for diagnosing brain tumors using deep learning techniques. First of all, the necessary theoretical background concerning fundamental concepts and architectures of neural networks are presented. Then, various methods for diagnosing brain tumors from related work of the past are investigated. In the proposed approach that is presented afterwards, the advantages of learning transfer are used to create five independent neural network models. In particular, the creation of one of them is based on the Mask R-CNN architecture and is used for brain tumor segmentation. The other four models are used for brain tumor classification and result from the combination of two VGGNet architecture's configurations, VGG-16 and VGG-19, with two learning transfer techniques. These five models form the basis of the two different approaches taken. Ιn the first approach, brain tumors are classified using the original data in the data set, four independent times, once for each classification model. In the second approach, tumors are first segmented and then classified using only the segmented part of the brain tumor. The procedure of the second approach is applied four independent times, once for each possible combination of the segmentation model with the classification models. The two approaches produce a total of eight different results, which elucidate the advantages and disadvantages of each option, as well as the prospect of deep learning becoming a key factor in the diagnosis of brain tumors.
Main subject category:
Technology - Computer science
Keywords:
Deep Learning, Convolutional Neural Networks, Segmentation, Classification, Brain Tumor Diagnosis