Skripsi

SEGMENTASI MULTI-CLASS BERBASIS TRANSFORMER PADA CITRA MEDIS MRI OTAK

---

Brain tumor segmentation from Magnetic Resonance Imaging (MRI) is crucial for accurate diagnosis and therapy planning. CNN architectures like U-Net are considered the gold standard, but they have intrinsic limitations in processing long-range spatial context globally. This study proposes the use of a Vision Transformer architecture, specifically UNETR, and compares its performance against a baseline 3D U-Net model for multi-class segmentation, targeting Enhancing Tumor (ET), Necrotic Core (NCR), and Peritumoral Edema (ED). The research utilized the public BraTS 2021 dataset containing four MRI modalities. The data was pre-processed using Random Spatial Crop and intensity normalization techniques. Model performance was evaluated using Dice Similarity Coefficient (DSC), Intersection over Union (IoU), and Hausdorff Distance 95 (HD95) metrics. Experimental results demonstrate that the 3D U-Net model slightly outperforms UNETR, achieving an average DSC of 0.8792 and an HD95 of 7.58 mm. Nevertheless, UNETR remains highly competitive, yielding an average DSC of 0.8649 and an HD95 of 10.30 mm. In conclusion, the local convolution operations and inductive bias of the CNN (3D U-Net) are still more efficient at mapping complex tumor boundaries compared to the Transformer's self-attention mechanism. The final model was implemented into a web-based software system using the Rational Unified Process (RUP) methodology.

---

Availability

Detail Information

Series Title

-

Call Number

T1959242026

Publisher

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

Collation

xviii, 53 hlm.; ilus.; 29 cm

Language

Indonesia

ISBN/ISSN

-

Classification

006.370 7

Content Type

Text

Media Type

-

Carrier Type

-

Edition

-

Subject(s)

Prodi Teknik Informatika
GoPay

Specific Detail Info

-

Statement of Responsibility

TUTI

Other version/related

File Attachment

Comments

---

You must be logged in to post a comment