SEGMENTASI SEMANTIK PEMBULUH DARAH CITRA RETINA MENGGUNAKAN ARSITEKTUR KOMBINASI U-NET, DEEPLABV3+, DAN ATTENTION GATE

SITORUS, DINA SUZZETE and Desiani, Anita and Resti, Yulia (2025) SEGMENTASI SEMANTIK PEMBULUH DARAH CITRA RETINA MENGGUNAKAN ARSITEKTUR KOMBINASI U-NET, DEEPLABV3+, DAN ATTENTION GATE. Undergraduate thesis, Sriwijaya University.

	Text RAMA_44201_08011382126082.pdf - Accepted Version Restricted to Repository staff only Available under License Creative Commons Public Domain Dedication. Download (1MB) | Request a copy
	Text RAMA_44201_08011382126082_TURNITIN.pdf - Accepted Version Restricted to Repository staff only Available under License Creative Commons Public Domain Dedication. Download (9MB) | Request a copy
	Text RAMA_44201_08011382126082_0011127702_0019077302_01_front_ref.pdf - Accepted Version Available under License Creative Commons Public Domain Dedication. Download (437kB)
	Text RAMA_44201_08011382126082_0011127702_0019077302_02.pdf - Accepted Version Restricted to Repository staff only Available under License Creative Commons Public Domain Dedication. Download (558kB) | Request a copy
	Text RAMA_44201_08011382126082_0011127702_0019077302_03.pdf - Accepted Version Restricted to Repository staff only Available under License Creative Commons Public Domain Dedication. Download (160kB) | Request a copy
	Text RAMA_44201_08011382126082_0011127702_0019077302_04.pdf - Accepted Version Restricted to Repository staff only Available under License Creative Commons Public Domain Dedication. Download (562kB) | Request a copy
	Text RAMA_44201_08011382126082_0011127702_0019077302_05.pdf - Accepted Version Restricted to Repository staff only Available under License Creative Commons Public Domain Dedication. Download (36kB) | Request a copy
	Text RAMA_44201_08011382126082_0011127702_0019077302_06_ref.pdf - Bibliography Available under License Creative Commons Public Domain Dedication. Download (138kB)

Abstract

The blood vessels in the retina are a system of blood vessels that function as deliverers of oxygen and nutrients. The blood vessels of the retina are divided into two parts, namely arteries and veins. Abnormalities in both blood vessels can indicate various diseases in the retina. Abnormal blood vessels can be analyzed in digital retinal images with image segmentation. This research performs semantic segmentation that produces three features, namely arterial, venous, and background blood vessels. The segmentation architecture used in this research is a combination of U-Net, DeepLabV3+, and Attention gate architectures. The use of DeepLabV3+ in the decoder aims to generate fewer parameters and extract features without reducing image resolution. The addition of Attention gate to the skip connection in the U-Net encoder aims to select important and unimportant features. The average performance results on accuracy, sensitivity, specificity, f1-score, and IoU are good enough in segmenting 96.32%, 79.11%, 91.96%, 80.97%, and 69.86%. Labeling results on the background label show that the accuracy, specificity, f1-score, and IoU values are good enough to segment above 95%, while the specificity is still at 77%. On arteries and veins labeling, the accuracy and specificity values are good enough to segment above 95%. However, the performance values on sensitivity, f1-score, and IoU are still below 95%. This is because there are very few features in arteries and veins to perform segmentation, so improvements are needed in this architecture to get sensitivity, f1-score, and IoU values above 95%.

Item Type:	Thesis (Undergraduate)
Uncontrolled Keywords:	Attention Gate, DeepLabV3+, Vessel Blood, U-Net, Segmentation
Subjects:	Q Science > QA Mathematics > QA75-76.95 Calculating machines > QA76.9.B45 Big data. Machine learning. Quantitative research. Metaheuristics.
Divisions:	08-Faculty of Mathematics and Natural Science > 44201-Mathematics (S1)
Depositing User:	Dina Suzzete Sitorus
Date Deposited:	30 Jan 2025 08:40
Last Modified:	30 Jan 2025 08:40
URI:	http://repository.unsri.ac.id/id/eprint/167323

Actions (login required)

View Item