Skripsi

SEGMENTASI TUMOR HATI PADA CITRA HASIL CT SCAN MENGGUNAKAN MODIFIKASI ARSITEKTUR U-NET, DENSENET, DAN ATTENTION GATE

---

Liver tumors are abnormal tissue growths in the liver that can be benign or malignant. Separating tumor tissue from healthy tissue in CT scan images can be done using segmentation techniques with Convolutional Neural Networks (CNN). The CNN architecture that can be used in image segmentation is U-Net. The encoder extracts important features from the image, but this process can cause the loss of detailed information or subtle features. The decoder serves to restore features to their original resolution but risks losing important information due to skip connections. This study modifies the U-Net architecture by adding DenseNet to the encoder to preserve fine image features and overcome the vanishing gradient problem, as well as Attention Gate to the decoder to filter spatial features, so that only relevant information is retained. Segmentation is performed with two labels, namely tumor and background. The liver tumor segmentation results obtained an accuracy of 97%, sensitivity of 93.3%, specificity of 97.2%, F1-Score of 72.1%, and IoU of 66.3%, indicating that the model is capable of accurately predicting tumor and background areas and has a good balance between sensitivity and specificity. This study proves that modifying U-Net with DenseNet and Attention Gate is effective in segmenting liver tumor images. Keywords: Liver Tumor, Segmentation, U-Net, DenseNet, Attention Gate

---

Availability

Detail Information

Series Title

-

Call Number

T1920902025

Publisher

Indralaya : Prodi Ilmu Matematika, Fakultas Matematika Dan Ilmu Pengetahuan Alam Universitas Sriwijaya., 2025

Collation

xviii, 151 hlm.; ilus.; 29 cm

Language

Indonesia

ISBN/ISSN

-

Classification

006.370 7

Content Type

Text

Media Type

-

Carrier Type

-

Edition

-

Subject(s)

Pengolahan Citra Digital
Prodi Ilmu Matematika

Specific Detail Info

-

Statement of Responsibility

TUTI

Other version/related

File Attachment

Comments

---

You must be logged in to post a comment