Skripsi

PERBANDINGAN ARSITEKTUR U-NET++ DAN U-NET UNTUK SEGMENTASI PADA CITRA X-RAY DADA

---

Chest X-ray (CXR) is a vital diagnostic modality for detecting lung diseases, yet manual interpretation is often hindered by low contrast and overlapping anatomical structures. Automatic lung segmentation serves as a crucial pre-processing step in Computer-Aided Diagnosis (CAD) systems. The standard U-Net architecture, despite its popularity, suffers from a "semantic gap" between encoder and decoder features, reducing precision at complex object boundaries. This study aims to implement and compare the performance of the U-Net++ architecture against the standard U-Net for lung segmentation tasks. U-Net++ introduces nested skip pathways and deep supervision innovations to bridge this semantic gap. The study utilized a combined dataset from Montgomery and Shenzhen, totaling 704 images with good class balance. Performance evaluation was conducted using Dice Similarity Coefficient (DSC) and Intersection over Union (IoU) metrics. Experimental results demonstrate that U-Net++ outperformed the standard U-Net, achieving a best DSC score of 96.6% and IoU of 93.6%, compared to U-Net's DSC of 96.1% and IoU of 92.5%. Although U-Net++ exhibited higher stability fluctuations during training due to the deep supervision mechanism, visual analysis confirmed that the model produced sharper and more accurate organ boundary delineation. This study concludes that U-Net++ effectively improves lung segmentation accuracy and holds potential for implementation in automated medical diagnosis systems.

---

Availability

Detail Information

Series Title

-

Call Number

T1897222025

Publisher

Indralaya : Prodi Teknik Informatika, Fakultas Ilmu Komputer Universitas Sriwijaya., 2025

Collation

xv, 106 hlm.; ilus.; tab.; 29 cm

Language

Indonesia

ISBN/ISSN

-

Classification

616.075 707

Content Type

Text

Media Type

-

Carrier Type

-

Edition

-

Subject(s)

Prodi Teknik Informatika
Diagnostik Pencitraan/X-Ray

Specific Detail Info

-

Statement of Responsibility

MI

Other version/related

File Attachment

Comments

---

You must be logged in to post a comment