Text

PENGEMBANGAN MODEL VISION TRANSFORMER UNTUK SEGMENTASI RUANG JANTUNG JANIN PADA FETAL ECHOCARDIOGRAPHY

---

Fetal heart chamber segmentation in fetal echocardiography (FE) is a big challenge in early diagnosis of congenital heart disease (CHD) due to low image contrast and morphological complexity. This study proposes the development of a vision transformer-based model through two main approaches: optimization of the YOLOv8-seg architecture and integration of the Segment Anything Model 2 (SAM2) hybrid model. Tests were conducted comprehensively using pathological datasets that included atrial septal defect (ASD), ventricular septal defect (VSD), and atrioventricular septal defect (AVSD) cases. The results show that the modification of the YOLOv8-seg architecture through an ablation study (Model M2), which involves the removal of feature map P5 and kernel optimization, successfully overcomes the overfitting problem on large-scale models. The YOLOv8x-seg M2 model recorded a drastic improvement in the mAP50-95 metric of 19.4% over the standard model, with a highly efficient inference time of 5.7 ms. On the other hand, the SAM2-YOLOv8s-det hybrid approach achieved the highest level of precision with an IoU score of 0.8789 and mAP50-95 of 0.7228. The attention mechanism in SAM2 proved superior in detecting subtle bulkhead defects in VSD and ASD cases compared to the pure convolution architecture. The main contribution of this research is the creation of two segmentation solution paths: (1) the M2 Model that is optimized for real-time screening needs on resource-constrained devices, and (2) the SAM2 Hybrid Model that provides high diagnostic accuracy for detailed analysis of septal abnormalities. These findings prove that the medical domain-specific adaptation of the transformer architecture can have a significant impact on improving the accuracy of automatic and objective early detection of fetal heart defects.

---

Availability

Detail Information

Series Title

-

Call Number

T1898342025

Publisher

Palembang : Prodi Doktor Ilmu Teknik, Fakultas Teknik Universitas Sriwijaya., 2025

Collation

xvii, 132 hlm.; ilus.; tab.; 29 cm.

Language

Indonesia

ISBN/ISSN

-

Classification

616.120 7

Content Type

Text

Media Type

-

Carrier Type

-

Edition

-

Subject(s)

Penyakit jantung bawaan
Prodi Doktor Ilmu Teknik

Specific Detail Info

-

Statement of Responsibility

MI

Other version/related

File Attachment

Comments

---

You must be logged in to post a comment