Perbaikan dan Augmentasi Citra Berbasis Transformer dan GAN Pada Segmentasi Citra Uji Silang Serasi Darah

Abimanyu, Ignatius Ida Bagus (2026) Perbaikan dan Augmentasi Citra Berbasis Transformer dan GAN Pada Segmentasi Citra Uji Silang Serasi Darah. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5025221189_Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (9MB) | Request a copy

Abstract

Uji silang serasi darah merupakan prosedur untuk memastikan kompatibilitas darah dalam proses donor darah, namun tipe pekerjaan repetitif seperti ini memiliki risiko human error. Penelitian ini bertujuan untuk mengembangkan model segmentasi citra medis guna membantu tenaga medis dalam menginterpretasikan hasil uji silang serasi. Metodologi yang digunakan berfokus pada tahap preprocessing dengan mengintegrasikan arsitektur Transformer untuk perbaikan kualitas citra dan arsitektur GAN untuk augmentasi data sintetis. Berdasarkan hasil pengujian, penggunaan teknik perbaikan citra Restormer memberikan peningkatan performa yang dengan kenaikan skor Intersection over Union sebesar 1,47 persen poin dari 0,8305 menjadi 0,8452 dan nilai precision mencapai 0,9124 karena kemampuannya mereduksi gangguan visual latar belakang secara efektif. Pada strategi augmentasi menggunakan GAN ditemukan tidak efektif karena gagal mereplikasi detail tekstur granular darah yang penting. Konfigurasi model tunggal terbaik dicapai oleh SwinUNETR, sementara metode ensemble mencapai performa tertinggi secara keseluruhan dengan IoU 0,8482. Jika dibandingkan dengan penelitian terdahulu, model yang diusulkan memiliki efisiensi lebih baik dengan waktu inferensi 0,1325 detik atau sekitar 7,4 kali lebih cepat dibandingkan metode sebelumnya.
=================================================================================================================================
Blood crossmatch testing is a procedure to ensure blood compatibility during the blood donation process; however, this type of work carries a risk of human error. This study aims to develop a medical image segmentation model to assist medical personnel in interpreting crossmatch test results. The methodology focuses on the preprocessing stage by integrating Transformer architectures for image quality enhancement and GAN architectures (pix2pix) for synthetic data augmentation. Based on experimental results, the use of the Restormer image enhancement technique provided a performance improvement with a 1.47 percentage point increase in IoU score from 0.8305 to 0.8452 and a precision value reaching 0.9124 due to its ability to effectively reduce background visual noise. The augmentation strategy using GAN was found to be ineffective as it failed to replicate essential granular blood texture details. The best single-model configuration was achieved by SwinUNETR, while the Ensemble method achieved the highest overall performance with an IoU of 0.8482. Compared to previous research, the proposed model demonstrates better efficiency with an inference time of 0.1325 seconds, or approximately 7.4 times faster than the previous method.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Augmentasi Citra, Generative Adversarial Networks (GAN), Perbaikan Citra, Restormer, Uji Silang Serasi, Blood Crossmatch, Generative Adversarial Networks (GAN), Image Augmentation, Image Enhancement , Restormer.
Subjects:	T Technology > T Technology (General) T Technology > T Technology (General) > T57.5 Data Processing
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55201-(S1) Undergraduate Thesis
Depositing User:	Ignatius Ida Bagus Abimanyu
Date Deposited:	22 Jan 2026 09:14
Last Modified:	22 Jan 2026 09:14
URI:	http://repository.its.ac.id/id/eprint/130101

Actions (login required)

View Item