Skripsi

SEGMENTASI JANTUNG ANAK MENGGUNAKAN VISION TRANSFORMER DAN SEGMENT ANYTHING MODEL

---

Congenital heart disease in children requires accurate early detection through ultrasonography (USG) imaging. This study aims to develop a deep learning–based system for view classification and hole segmentation in pediatric cardiac images. A Vision Transformer (ViT) model was employed to classify five cardiac views (4CH, 5CH, LA, SA, and SUB), while segmentation was performed using YOLO11-seg and Segment Anything Model 2 (SAM 2) with a bounding box–based prompt approach. The dataset consisted of 800 USG videos that were converted into frames and divided into training, validation, and unseen sets. Classification performance was evaluated using accuracy, precision, and recall metrics, while segmentation performance was assessed using Intersection over Union (IoU) and Dice coefficient. The results show that ViT-L/32 achieved the best classification performance with an accuracy of 92.20%. For the segmentation task, the best YOLO11-seg model achieved an IoU of 36.28%, and the combination of YOLO11-seg + SAM 2 produced an IoU of 48.53%, demonstrating the model’s capability to precisely identify hole regions. This study highlights the potential application of Vision Transformer and SAM in supporting computer-aided diagnosis systems for detecting congenital heart abnormalities in children.

---

Availability

Detail Information

Series Title

-

Call Number

T1951032026

Publisher

Indralaya : Prodi Sistem Komputer, Fakultas Ilmu Komputer Universitas Sriwijaya., 2026

Collation

xvii, 66 hlm.; ilus.; 29 cm

Language

Indonesia

ISBN/ISSN

-

Classification

006.310 7

Content Type

Text

Media Type

unmediated

Carrier Type

-

Edition

-

Subject(s)

Prodi Sistem Komputer
Pembelajaran Mesin

Specific Detail Info

-

Statement of Responsibility

KA

Other version/related

File Attachment

Comments

---

You must be logged in to post a comment