Skripsi

IMAGE CAPTIONING UNTUK GENERASI PRA-DIAGNOSIS RADIOLOGI MENGGUNAKAN ARSITEKTUR TRANSFORMER

---

General vision-language models often face domain shift issues when applied to medical imaging. This study compares the performance of a general model (BLIP) against a medically pre-trained model (MedBLIP) for Chest X-ray reporting using the IU X-Ray dataset. Through quantitative metrics and expert pulmonologist validation, MedBLIP consistently outperformed BLIP, achieving a BLEU-4 of 0.0789, ROUGE-L of 0.3199, and BERTScore of 0.7017. Expert validation confirmed MedBLIP’s superiority with a higher average clinical accuracy score (3.50 vs. 3.40). While both models struggled with dataset imbalance, MedBLIP demonstrated better capability in capturing specific pathological terminology compared to BLIP’s generic narratives. Conclusively, medical domain-specific pretraining significantly enhances the semantic quality and clinical relevance of automated pre-diagnosis reports.

---

Availability

Detail Information

Series Title

-

Call Number

T1900742025

Publisher

Indralaya : Prodi Teknik Informatika, Fakultas Ilmu Komputer Universitas Sriwijaya., 2025

Collation

xv, 122 hlm.; ilus.; tab.; 29 cm

Language

Indonesia

ISBN/ISSN

-

Classification

616.075 407

Content Type

Text

Media Type

-

Carrier Type

-

Edition

-

Subject(s)

Prodi Teknik Informatika
Diagnostik Terkomputerisasi

Specific Detail Info

-

Statement of Responsibility

MI

Other version/related

File Attachment

Comments

---

You must be logged in to post a comment