SEGMENTASI PEMBULUH DARAH PADA CITRA RETINA MENGGUNAKAN METODE CONVOLUTIONAL NEURAL NETWORK VV-NET

AGUSTINA, SINTA BELLA and Erwin, Erwin and Desiani, Anita (2024) SEGMENTASI PEMBULUH DARAH PADA CITRA RETINA MENGGUNAKAN METODE CONVOLUTIONAL NEURAL NETWORK VV-NET. Masters thesis, Sriwijaya University.

[thumbnail of RAMA_55101_09012682226010.pdf] Text
RAMA_55101_09012682226010.pdf - Accepted Version
Restricted to Repository staff only
Available under License Creative Commons Public Domain Dedication.

Download (10MB) | Request a copy
[thumbnail of RAMA_55101_09012682226010_TURNITIN.pdf] Text
RAMA_55101_09012682226010_TURNITIN.pdf - Accepted Version
Restricted to Repository staff only
Available under License Creative Commons Public Domain Dedication.

Download (7MB) | Request a copy
[thumbnail of RAMA_55101_09012682226010_0029017101_0011127702_01_front_ref.pdf] Text
RAMA_55101_09012682226010_0029017101_0011127702_01_front_ref.pdf - Accepted Version
Available under License Creative Commons Public Domain Dedication.

Download (1MB)
[thumbnail of RAMA_55101_09012682226010_0029017101_0011127702_02.pdf] Text
RAMA_55101_09012682226010_0029017101_0011127702_02.pdf - Accepted Version
Restricted to Repository staff only
Available under License Creative Commons Public Domain Dedication.

Download (442kB) | Request a copy
[thumbnail of RAMA_55101_09012682226010_0029017101_0011127702_03.pdf] Text
RAMA_55101_09012682226010_0029017101_0011127702_03.pdf - Accepted Version
Restricted to Repository staff only
Available under License Creative Commons Public Domain Dedication.

Download (288kB) | Request a copy
[thumbnail of RAMA_55101_09012682226010_0029017101_0011127702_04.pdf] Text
RAMA_55101_09012682226010_0029017101_0011127702_04.pdf - Accepted Version
Restricted to Repository staff only
Available under License Creative Commons Public Domain Dedication.

Download (256kB) | Request a copy
[thumbnail of RAMA_55101_09012682226010_0029017101_0011127702_05.pdf] Text
RAMA_55101_09012682226010_0029017101_0011127702_05.pdf - Accepted Version
Restricted to Repository staff only
Available under License Creative Commons Public Domain Dedication.

Download (45kB) | Request a copy
[thumbnail of RAMA_55101_09012682226010_0029017101_0011127702_06_ref.pdf] Text
RAMA_55101_09012682226010_0029017101_0011127702_06_ref.pdf - Bibliography
Restricted to Repository staff only
Available under License Creative Commons Public Domain Dedication.

Download (200kB) | Request a copy
[thumbnail of RAMA_55101_09012682226010_0029017101_0011127702_07_lamp.pdf] Text
RAMA_55101_09012682226010_0029017101_0011127702_07_lamp.pdf - Accepted Version
Restricted to Repository staff only
Available under License Creative Commons Public Domain Dedication.

Download (7MB) | Request a copy

Abstract

The U-Net architecture has a fairly deep network. The addition of layers in the U-Net architecture network can increase the complexity of the U-Net network which can affect the training time to be longer and parameter enlargement. This study modifies U-Net by reducing the complexity of U-Net by removing the bridge part of U-Net. The removal of the bridge in U-Net is known as V-Net architecture. The removal of the bridge has the risk of underfitting. To avoid the risk of underfitting, a modification of V-Net is proposed by performing V-Net twice for blood vessel segmentation. The application of V-Net twice is referred to as VV-Net architecture. The first V-Net is used for feature extraction and the second V-Net is used to improve feature extraction so as to produce better segmentation. This study aims to determine the performance evaluation results of the VV-Net architecture. The evaluation measures used are accuracy, sensitivity, precision and Jaccard score. Tests were conducted on the DRIVE, STARE, and CHASEDB_1 datasets. The measurement results of blood vessel segmentation using VV-net on the DRIVE dataset resulted in accuracy 96.27%, sensitivity 84.38%, precision 75.95%, and Jaccard score 66.28%. On the STARE dataset, the accuracy result is 96.58%, sensitivity 82.78%, precission 76.73%, and Jaccard score 65.38%. Meanwhile, the CHASEDB_1 dataset resulted in 97.04% accuracy, 83.55% sensitivity, 76.72% precission, and 66.40% Jaccard score. Based on these results, it shows that the proposed VV-Net architecture is very good in segmenting blood vessels, indicated by accuracy values above 90%, sensitivity above 80%, and precision above 70%. The Jaccard score value is still below 70%, indicating that the proposed architecture is quite good at detecting faint blood vessel regions. Since the Jaccard score value is still below 70%, the focus of further research is to make improvements to the proposed architecture to increase the Jaccard score value.

Item Type: Thesis (Masters)
Uncontrolled Keywords: Retinal; Convolutional Neural Network; U-Net; V-Net; Segmentation Blood Vessels, VV-Net.
Subjects: T Technology > TR Photography > TR624-835 Applied photography Including artistic, commercial, medical photography, photocopying processes
Divisions: 09-Faculty of Computer Science > 55101-Informatics (S2)
Depositing User: Sinta Bella Agustina
Date Deposited: 22 May 2024 06:50
Last Modified: 22 May 2024 06:50
URI: http://repository.unsri.ac.id/id/eprint/145161

Actions (login required)

View Item View Item