Segmentasi Otomatis Menggunakan Model Shifted Window U-Net (Swin-Unet) sebagai Basis Analisis Kuantitatif Citra T2 Magnetic Resonance Imaging (MRI)

Fatikhah, Irene Nurbaiti (2026) Segmentasi Otomatis Menggunakan Model Shifted Window U-Net (Swin-Unet) sebagai Basis Analisis Kuantitatif Citra T2 Magnetic Resonance Imaging (MRI). Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5001221021-Undergraduate_thesis.pdf - Accepted Version Restricted to Repository staff only Download (2MB) | Request a copy

Abstract

Pengukuran intensitas sinyal pada *Region of Interest* (ROI) dalam citra *Magnetic Resonance Imaging* (MRI) umumnya masih dilakukan secara manual menggunakan perangkat lunak seperti ImageJ. Proses ini memerlukan waktu yang relatif lama dan rentan terhadap variasi antar-pengamat. Penelitian ini bertujuan mengembangkan sistem segmentasi otomatis yang memanfaatkan arsitektur Swin-Unet berbasis Swin Transformer untuk mendukung proses pengukuran ROI pada citra MRI. Data yang digunakan terdiri atas 112 file DICOM dengan *Repetition Time* (TR) konstan sebesar 1000 ms dan delapan variasi *Echo Time* (TE), yaitu 16, 25, 50, 75, 100, 150, 200, dan 250 ms. Model dilatih selama 100 *epoch* menggunakan optimizer Adam dan fungsi *loss* binary cross-entropy. Hasil evaluasi menunjukkan nilai Dice Coefficient (DSC) sebesar 0,9440, Intersection over Union (IoU) sebesar 0,8940, ROC-AUC sebesar 0,9996, dan Dice Loss sebesar 0,0553. Model berhasil mendeteksi 21 ROI pada seluruh 112 irisan citra serta mengekstraksi 2.352 data ROI secara otomatis. Hasil analisis menunjukkan bahwa variasi TE memengaruhi nilai intensitas sinyal ROI, dengan kecenderungan intensitas menurun seiring meningkatnya nilai TE. Penurunan terbesar terjadi pada TE 75 ms, yaitu sekitar 15,60% dibandingkan TE 16 ms, yang sesuai dengan teori relaksasi T2. Peningkatan kembali intensitas pada rentang TE 150–250 ms mengindikasikan karakteristik relaksasi T2 multikomponen pada fantom yang terdiri atas beberapa material dengan nilai T2 yang berbeda.
==================================================================================================================================
Signal intensity measurement within the *Region of Interest* (ROI) in *Magnetic Resonance Imaging* (MRI) is generally still performed manually using software such as ImageJ. This process is time-consuming and prone to inter-observer variability. This study aims to develop an automated segmentation system utilizing a Swin-Unet architecture based on the Swin Transformer to support ROI measurement in MRI images. The dataset consists of 112 DICOM files with a constant *Repetition Time* (TR) of 1000 ms and eight *Echo Time* (TE) variations, namely 16, 25, 50, 75, 100, 150, 200, and 250 ms. The model was trained for 100 epochs using the Adam optimizer and the binary cross-entropy loss function. The evaluation results yielded a Dice Coefficient (DSC) of 0.9440, an Intersection over Union (IoU) of 0.8940, an ROC-AUC of 0.9996, and a Dice Loss of 0.0553. The model successfully detected 21 ROIs across all 112 image slices and automatically extracted 2,352 ROI data points. The analysis indicates that TE variation influences ROI signal intensity, with intensity generally decreasing as TE increases. The greatest reduction occurred at a TE of 75 ms, with an approximately 15.60% decrease compared to TE 16 ms, consistent with T2 relaxation theory. The subsequent increase in signal intensity within the TE range of 150–250 ms suggests multi-component T2 relaxation characteristics in the phantom, which consists of several materials with different T2 values.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Deep Learning, Intensitas Sinyal, MRI, Segmentasi Otomatis, Swin-Unet., Automatic Segmentation, Deep Learning, MRI, Signal Intensity, Swin-Unet.
Subjects:	Q Science > QC Physics R Medicine > R Medicine (General) > R858 Deep Learning R Medicine > RC Internal medicine > RC78.7.N83 Magnetic resonance imaging.
Divisions:	Faculty of Science and Data Analytics (SCIENTICS) > Physics > 45201-(S1) Undergraduate Thesis
Depositing User:	Irene Nurbaiti Fatikhah
Date Deposited:	01 Aug 2026 02:19
Last Modified:	01 Aug 2026 02:19
URI:	http://repository.its.ac.id/id/eprint/141430

Actions (login required)

View Item