Segmentasi Tumor Hati Menggunakan Arsitektur 2,5D CNN Berbasis Residual U-Net dan VGG16 U-Net

Simamora, Yoel (2026) Segmentasi Tumor Hati Menggunakan Arsitektur 2,5D CNN Berbasis Residual U-Net dan VGG16 U-Net. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5003221065-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (4MB) | Request a copy

Abstract

Kanker hati merupakan salah satu penyakit dengan tingkat mortalitas tinggi yang sering terdeteksi pada stadium lanjut, sehingga deteksi dini menjadi sangat penting untuk meningkatkan peluang keberhasilan terapi. Segmentasi tumor hati secara otomatis pada citra CT scan dapat mendukung proses diagnosis dan perencanaan terapi secara lebih cepat dan akurat. Penelitian ini mengusulkan pendekatan segmentasi tumor hati berbasis 2,5D Convolutional Neural Network (CNN) dengan dua arsitektur deep learning, yaitu Residual U-Net dan VGG16 U-Net. Data yang digunakan berasal dari Liver Tumor Segmentation Challenge (LiTS) dan melalui tahapan preprocessing, segmentasi dua tahap (segmentasi hati untuk pembentukan Region of Interest dan segmentasi tumor hati), serta post-processing berbasis volume 3D untuk menjaga kontinuitas hasil segmentasi dan mengurangi false positive. Evaluasi kinerja model dilakukan menggunakan metrik Dice Similarity Coefficient (DSC), Intersection over Union (IoU), dan Correct Classification Ratio (CCR). Hasil penelitian menunjukkan bahwa penerapan post-processing secara signifikan meningkatkan nilai DSC dan IoU pada kedua arsitektur. Pada data testing setelah post-processing, model Residual U-Net menghasilkan nilai DSC sebesar 0,7088 dan IoU sebesar 0,6970, sedangkan model VGG16 U-Net memperoleh nilai DSC sebesar 0,6986 dan IoU sebesar 0,6860. Meskipun kedua model menunjukkan performa yang sebanding, Residual U-Net memberikan hasil segmentasi yang sedikit lebih unggul dan konsisten secara spasial. Berdasarkan hasil tersebut, dapat disimpulkan bahwa pendekatan 2,5D CNN yang dikombinasikan dengan arsitektur deep learning dan post-processing berbasis informasi spasial efektif dalam meningkatkan kualitas segmentasi tumor hati. Penelitian ini diharapkan dapat memberikan kontribusi dalam pengembangan metode segmentasi citra medis yang efisien dan akurat, serta berpotensi mendukung aplikasi klinis dan penelitian lanjutan dalam deteksi dan analisis tumor hati.
==================================================================================================================================
Liver cancer is one of the leading causes of cancer-related mortality and is often diagnosed at an advanced stage, making early detection essential for improving treatment outcomes. Automatic liver tumor segmentation on computed tomography (CT) images can support the diagnostic process and treatment planning more quickly and accurately. This study proposes a liver tumor segmentation approach based on a 2,5D Convolutional Neural Network (CNN) using two deep learning architectures, namely Residual U-Net and VGG16 U-Net. The dataset was obtained from the Liver Tumor Segmentation Challenge (LiTS) and underwent several stages, including preprocessing, a two-stage segmentation process (liver segmentation to obtain the Region of Interest followed by liver tumor segmentation), and 3D volume-based post-processing to preserve spatial continuity and reduce false positives. Model performance was evaluated using the Dice Similarity Coefficient (DSC), Intersection over Union (IoU), and Correct Classification Ratio (CCR). The results showed that post-processing significantly improved the DSC and IoU values for both architectures. On the testing dataset after post-processing, the Residual U-Net achieved a DSC of 0,7088 and an IoU of 0,6970, while the VGG16 U-Net obtained a DSC of 0,6986 and an IoU of 0,6860. Although both models demonstrated comparable performance, the Residual U-Net provided slightly better and more spatially consistent segmentation results. These findings indicate that a 2.5D CNN approach combined with deep learning architectures and spatially informed post-processing is effective in improving liver tumor segmentation quality. This study is expected to contribute to the development of efficient and accurate medical image segmentation methods and has the potential to support clinical applications and further research in liver tumor detection and analysis.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Segmentation, CNN, Liver Tumor, 2,5D, Residual UNet, VGG16 UNet Segmentasi, CNN, Tumor Hati, 2,5D, Residual UNet, VGG16 UNet
Subjects:	H Social Sciences > HA Statistics Q Science > QA Mathematics > QA76.87 Neural networks (Computer Science) R Medicine > R Medicine (General) > R858 Deep Learning T Technology > TA Engineering (General). Civil engineering (General) > TA1637 Image processing--Digital techniques. Image analysis--Data processing.
Divisions:	Faculty of Science and Data Analytics (SCIENTICS) > Statistics > 49201-(S1) Undergraduate Thesis
Depositing User:	Yoel Prawira Simamora
Date Deposited:	27 Jan 2026 01:27
Last Modified:	27 Jan 2026 01:27
URI:	http://repository.its.ac.id/id/eprint/130637

Actions (login required)

View Item