Text

SEGMENTASI TUMOR OTAK PADA CITRA HASIL MAGNETIC RESONANCE IMAGING OTAK MENGGUNAKAN ARSITEKTUR 3D DOUBLE V-NET DENGAN ATTENTION GATE

---

Penilaian

---

Brain tumors are disorders caused by abnormal cell growth in the brain. Early detection is important to recognize tumor patterns in Magnetic Resonance Imaging (MRI) images. Early detection is done by separating tumor cell details through segmentation using the Convolutional Neural Network (CNN) method. This study proposes a 3D Double V-Net architecture by incorporating attention gates in each decoder within the skip connections. The 3D Double V-Net is a modified version of the 3D Double U-Net, where the bridge component is removed. This removal aims to reduce the number of parameters that could lead to overfitting. Attention gates are introduced in each decoder to refine feature selection from the skip connections. The proposed architecture is applied to brain tumor segmentation in three-dimensional MRI scans, which contain length, width, and depth information. The segmentation process is conducted on four labels: background, non-enhancing tumor, peritumoral edema, and enhancing tumor. The model's performance is evaluated using accuracy, sensitivity, specificity, IoU and F1-score. The implementation of the proposed architecture resulted in average performance on accuracy, sensitivity, specificity, IoU, and f1-score of 99.4%, 92.35%, 97.78%, 86.6%, and 92.67%. The model’s performance on the background label showed excellent results, with accuracy, sensitivity, specificity, IoU, and F1-score above 90%. The non-enhancing tumor label also demonstrated excellent performance, with accuracy, sensitivity, specificity, and F1-score above 90%, although the IoU value remained below 90%. The peritumoral edema and enhancing tumor labels showed excellent performance with accuracy, specificity, and F1-score above 90%, but sensitivity and IoU were still below 90%. This is due to the fact that the size of the features in the label enhancing tumor and peritumoral edema is relatively small and the edge boundaries are not clear, making it difficult to recognize. Based on this, architectural improvements are needed to increase sensitivity and IoU values above 90%.

---

Availability

Inventory Code	Barcode	Call Number	Location	Status
2507001934	T169733	T1697332025	Central Library (Reference)	Available but not for loan - Not for Loan

Detail Information

Series Title

-

Call Number

T1697332025

Publisher

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

Collation

x, 72 hlm.: Ilus., tab.; 29 cm

Language

Indonesia

ISBN/ISSN

-

Classification

510.07

Content Type

-

Media Type

-

Carrier Type

-

Edition

-

Subject(s)

Matematika

Specific Detail Info

-

Statement of Responsibility

EM

Other version/related

File Attachment

Comments

---

You must be logged in to post a comment