Model Segmentasi Otomatis Region of Interest pada Citra Relaksasi Spin-Spin (T2) Magnetic Resonance Imaging (MRI) Fantom Gel Berbasis U-Net

Khasanah, Aisyah Nur (2026) Model Segmentasi Otomatis Region of Interest pada Citra Relaksasi Spin-Spin (T2) Magnetic Resonance Imaging (MRI) Fantom Gel Berbasis U-Net. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5001221005 - Undergaduate_thesis.pdf - Accepted Version Restricted to Repository staff only Download (2MB) | Request a copy

Abstract

Magnetic Resonance Imaging (MRI) merupakan salah satu modalitas pencitraan yang memanfaatkan medan magnet dan gelombang radio untuk dapat mencitrakan organ tubuh manusia. Teknologi deep learning dapat dimanfaatkan untuk membantu pengolahan citra MRI. Untuk itu dibuatlah model segmentasi otomatis berbasis U-Net yang bertujuan untuk merancang, mengimplementasikan, dan mengevaluasi model segmentasi otomatis ROI menggunakan U-Net. Model dilatih menggunakan 160 citra MRI T2 fantom gel yang terbagi menjadi 96 citra sebagai data latih, 32 citra sebagai data validasi, dan 32 citra sebagai data uji. Citra tersebut merupakan hasil dari proses augmentasi yang berfungsi untuk memperkaya dataset pelatihan model. Evaluasi internal model diperoleh nilai akurasi sebesar 99,07%, presisi sebesar 93,77%, recall sebesar 99,94%, specificity sebesar 98,93%, F1-Score sebesar 96,75%, IoU sebesar 93,71%, dan AUC sebesar 99,98%. Berdasarkan hasil evaluasi tersebut model dapat digunakan untuk segmentasi ROI secara otomatis. Citra yang diuji adalah hasil pencitraan MRI dengan parameter yang digunakan yaitu TR pada 1000 ms dengan variasi nilai TE pada 16, 25, 50, 75, 100, 150, 200, dan 250 ms. Setiap variasi TE, citra fantom gel diiris menjadi 18 irisan. 18 irisan citra dapat disegmentasikan sekaligus dan dihitung intensitas sinyal pada ROI hasil segmentasi sehingga dapat mengefisiensi waktu. Nilai intensitas sinyal citra hasil segmentasi otomatis dibandingkan dengan intensitas sinyal ketika diukur secara manual menggunakan Image-J diperoleh tingkat kesesuaiannya melalui koefisien determinasi yang dihasilkan yaitu 0,9998.
===================================================================================================================================
Magnetic Resonance Imaging (MRI) is one of the imaging modalities that utilizes magnetic fields and radio waves to image human organs. Deep learning technology can be used to assist in the processing of MRI images. To that end, an automatic segmentation model based on U-Net was developed, aiming to design, implement, and evaluate an automatic ROI segmentation model using U-Net. The model was trained using 160 T2 MRI images of gel phantoms, divided into 96 images for training data, 32 images for validation data, and 32 images for test data. These images are the result of an augmentation process that serves to enrich the model's training dataset. Internal evaluation of the model yielded an accuracy of 99.07%, precision of 93.77%, recall of 99.94%, specificity of 98.93%, F1-Score of 96.75%, IoU of 93.71%, and AUC of 99.98%. Based on these evaluation results, the model can be used for automatic ROI segmentation. The images tested were the results of MRI imaging with parameters set at a TR of 1000 ms with varying TE values of 16, 25, 50, 75, 100, 150, 200, and 250 ms. For each TE variation, the gel phantom image was sliced into 18 slices. All 18 image slices can be segmented simultaneously, and the signal intensity within the segmented ROI can be calculated, thereby improving efficiency. The signal intensity values of the automatically segmented images were compared with the signal intensity measured manually using Image-J, and the level of agreement, obtained through the resulting coefficient of determination, was 0.9998.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Citra MRI, Deep learning, Segmentasi Otomatis, U-Net., Automatic Segmentation, Deep learning, MRI Image, U-Net.
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:	Aisyah Nur Khasanah
Date Deposited:	01 Aug 2026 02:14
Last Modified:	01 Aug 2026 02:14
URI:	http://repository.its.ac.id/id/eprint/141427

Actions (login required)

View Item