Pengembangan Sistem Node Programming Untuk Segmentasi Citra Medis Menggunakan Arsitektur Mikroservis Asinkron Berbasis RabbitMQ

Sanubari, Chalwat (2026) Pengembangan Sistem Node Programming Untuk Segmentasi Citra Medis Menggunakan Arsitektur Mikroservis Asinkron Berbasis RabbitMQ. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5025221259-Undergraduate_Thesis.pdf Restricted to Repository staff only Download (8MB) | Request a copy

Abstract

Segmentasi citra medis memainkan peran krusial dalam diagnosis klinis, terutama melalui pemanfaatan arsitektur Deep Learning seperti U-Net. Namun, ekosistem perangkat lunak yang ada saat ini masih terfragmentasi, di mana proses pelabelan anotasi dan pelatihan model berjalan secara terpisah. Selain itu, tingginya hambatan dalam penulisan kode pemrograman tingkat rendah (low-level coding) serta besarnya beban komputasi untuk pelatihan model sering kali menjadi kendala operasional. Oleh karena itu, diperlukan sebuah sistem terpadu yang mampu memfasilitasi siklus umpan balik (feedback loop) dari proses anotasi hingga inferensi secara mulus tanpa memicu kelumpuhan infrastruktur akibat beban berlebih. Penelitian ini mengusulkan rancang bangun platform terintegrasi menggunakan paradigma Visual Node Programming untuk menyederhanakan konfigurasi model segmentasi citra medis tanpa kode. Arsitektur sistem dikembangkan menggunakan pola Three-Tier berbasis layanan mikro (microservices). Guna menangani beban komputasi pelatihan yang berat, arsitektur memisahkan logika antarmuka pada API Golang dengan unit komputasi pada Python Worker secara asinkron. Sistem ini mengimplementasikan message broker RabbitMQ untuk distribusi beban, sehingga antrean tugas pelatihan dapat ditahan secara aman di luar memori pekerja. Pengujian fungsional divalidasi secara end-to-end menggunakan Brain Tumor Segmentation Dataset untuk tugas manajemen anotasi dan inferensi. Sementara itu, evaluasi non-fungsional dianalisis secara matematis menggunakan Teori Antrean (Hukum Little). Hasil pengujian kinerja menunjukkan bahwa arsitektur berhasil meredam lonjakan beban ekstrem hingga 100 Pengguna Virtual tanpa mengalami saturasi memori (Out-Of-Memory). Selain itu, sistem menunjukkan efisiensi skalabilitas beban secara linear, di mana penambahan kapasitas komputasi dari 1 menjadi 4 worker terbukti mampu mendongkrak laju penyelesaian tugas dan memangkas waktu tunggu antrean (queuing latency) hingga 77%.
================================================================================================================================
Medical image segmentation plays a crucial role in clinical diagnosis, particularly through the utilization of Deep Learning architectures such as U-Net. However, the current software ecosystem remains fragmented, with the annotation labeling process and model training operating separately. Furthermore, the high barrier of low-level code writing and the heavy computational load required for model training often pose operational challenges. Therefore, an integrated system capable of facilitating a seamless feedback loop from annotation to inference, without triggering infrastructure collapse due to overload, is urgently needed. This study proposes the design and development of an integrated platform using the Visual Node Programming paradigm to simplify the configuration of medical image segmentation models in a no-code manner. The system architecture is developed using a microservices-based Three-Tier pattern. To handle heavy computational training loads, the architecture asynchronously separates the interface logic in the Golang API from the computational units in the Python Worker. The system implements the RabbitMQ message broker to achieve dynamic load balancing, ensuring that training task queues are safely held outside worker memory. Functional testing was validated end-to-end using the Brain Tumor Segmentation Dataset for annotation management and inference tasks. Meanwhile, non-functional evaluations were mathematically analyzed using Queueing Theory (Little's Law). Performance results demonstrate that the architecture successfully mitigates extreme load surges of up to 100 Virtual Users without experiencing memory saturation (Out-Of-Memory). Additionally, the system exhibits linear load scalability efficiency, where increasing computational capacity from 1 to 4 workers proved capable of boosting task throughput and cutting queuing latency by up to 77%.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Anotasi Citra, Node Programming, Mikroservis, RabbitMQ, Segmentasi Citra Medis, U-Net, Image Annotation, Medical Image Segmentation, Microservices, Node Programming, RabbitMQ, U-Net
Subjects:	T Technology > TA Engineering (General). Civil engineering (General) > TA1637 Image processing--Digital techniques. Image analysis--Data processing.
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55201-(S1) Undergraduate Thesis
Depositing User:	Chalwat Sanubari
Date Deposited:	28 Jul 2026 06:26
Last Modified:	28 Jul 2026 06:26
URI:	http://repository.its.ac.id/id/eprint/138617

Actions (login required)

View Item