Skripsi

SEGMENTASI CITRA PARU-PARU MENGGUNAKAN ARSITEKTUR CUSTOM U-NET

---

Manual analysis of COVID-19 chest CT scans is often time-consuming and prone to subjectivity, thereby hindering diagnostic efficiency. Therefore, this study aims to develop an automatic segmentation system using a Custom U-Net architecture with a pre-trained DenseNet-169 encoder to accurately map infection areas, including Ground Glass Opacity (GGO), Consolidation, and Pleural Effusion classes. This study utilizes a public dataset from Medical Segmentation and evaluates various training scenarios through variations in batch size, learning rate, and optimizer, while comparing it with the standard Vanilla U-Net architecture as a baseline. The test results show that the best configuration for Custom U-Net is achieved using the Adam optimizer with a batch size of 8 and differential learning rates, yielding a Mean Dice score of 0.4883 and a Mean IoU score of 0.3451. Comparative analysis proves that the proposed architecture successfully outperforms the standard Vanilla U-Net model, which only achieved a Mean Dice of 0.4398. Furthermore, the feature extraction mechanism in the DenseNet-169 encoder significantly improves the model's localization ability on the most challenging minority class, demonstrated by the increase in the Dice score for Pleural Effusion from 0.1999 in the baseline model to 0.2684. Overall, this architectural modification provides superior visual representation and segmentation accuracy to support medical analysis. This intelligent system innovation is expected to contribute directly to the achievement of the 3rd Sustainable Development Goal (SDG 3), which is to ensure healthy lives and promote well-being for all at all ages by improving the efficiency of healthcare services.

---

Availability

Detail Information

Series Title

-

Call Number

T1961082026

Publisher

Indralaya : Prodi Teknik Informatika, Fakultas Ilmu Komputer Universitas Sriwijaya., 2026

Collation

xx, V-20 hlm.; ilus.; 29 cm

Language

Indonesia

ISBN/ISSN

-

Classification

006.312 07

Content Type

Text

Media Type

unmediated

Carrier Type

-

Edition

-

Subject(s)

Machine Learning
Prodi Teknik Informatika

Specific Detail Info

-

Statement of Responsibility

KA

Other version/related

File Attachment

Comments

---

You must be logged in to post a comment