Pengaruh Reorientasi Anterior Commisure – Posterior Commisure dan Tuning Hyperparameter Untuk Kasus Segmentasi 3D Aneurisma di Otak

Ryan, Richard (2025) Pengaruh Reorientasi Anterior Commisure – Posterior Commisure dan Tuning Hyperparameter Untuk Kasus Segmentasi 3D Aneurisma di Otak. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5025211141-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (6MB) | Request a copy

Abstract

Penelitian ini mengkaji dampak reorientasi Anterior Commissure-Posterior Commissure (AC-PC) dan konfigurasi fungsi loss serta optimizer terhadap akurasi segmentasi aneurisma 3D pada Magnetic Resonance Angiography (MRA) otak menggunakan model UNetR. Segmentasi yang akurat sangat penting untuk mendiagnosis kondisi yang mengancam jiwa, seperti stroke hemoragik. Landmark AC-PC memberikan titik referensi yang konsisten untuk penyelarasan yang lebih baik. Terdapat dua metrik utama yang digunakan dalam penelitian ini, Dice Similarity Coefficient (DSC) dan Hausdorff Distance (HD). Dua eksperimen prapemrosesan dilakukan: (1) mengubah ukuran menjadi (256, 256, 256) dan (2) mengubah ukuran menjadi (256, 256, 256) diikuti dengan reorientasi berbasis AC-PC. Akan dilakukan pula pengujian beberapa konfigurasi fungsi loss: hanya Dice Loss dan berbagai kombinasi Dice Loss dan Binary Cross Entropy (BCE) Loss. Selain itu, juga akan dilakukan pengujian 6 jenis optimizer yaitu AdamW, Adam, Adagrad, Adadelta, RMSProp, dan SGD. Pengujian akan menggunakan pengujian Wilcoxon yang dapat menguji 2 sampel yang berhubungan namun tidak terdistribusi normal. Ditemukan bahwa reorientasi memberikan dampak negatif namun tidak signifikan terhadap performa segmentasi. Konfigurasi loss dan optimizer terbaik adalah dengan menggunakan gabungan Dice Loss dan BCE Loss dengan rasio 50:50 serta menggunakan optimizer SGD. Model terbaik berhasil mendapatkan rata-rata DSC 0,286 dan HD 51,678. Dalam beberapa kasus khusus, model bahkan berhasil mendapatkan skor DSC 0,9092 dan HD 1,4142.
=================================================================================================================================
This study examines the impact of Anterior Commissure-Posterior Commissure (AC-PC) reorientation and hyperaparameter tuning on 3D aneurysm segmentation accuracy in brain MRA using the UNetR model. Accurate segmentation is vital for diagnosing life-threatening conditions like hemorrhagic stroke. AC-PC landmarks provide consistent reference points for better alignment. Two preprocessing experiments were conducted: (1) reshaping to (256, 256, 256) and (2) reshaping to (256, 256, 256) followed by AC-PC-based rotation. Several loss function setups were tested: Dice Loss alone and a variety of combination of Dice Loss and Binary Cross Entropy (BCE) Loss. Aside from that, six kinds of optimizers will also be tested, which includes AdamW, Adam, Adagrad, Adadelta, RMSProp, and SGD. Testing will be using the Wilcoxon Signed-rank test which can test 2 related samples but not under normal distribution. It was found that reorientation gave a worse result but not a significant one for the segmentation performance. The best configuration found for loss function and optimizer is a mix of Dice Loss and BCE Loss with ratio 50:50 with SGD as optimizer. The best model achieved an average DSC of 0.286 and HD 51.678. In some special cases, the models even managed to get a DSC of 0.9092 and HD 1.4142

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Brain Aneurysm, Deep Learning, Medical Image Preprocessing, Semantic Segmentation, Aneurisma Otak, Deep Learning, Preprocessing Citra Medis, Segmentasi Semantik
Subjects:	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:	Richard Ryan
Date Deposited:	14 Jul 2025 01:55
Last Modified:	14 Jul 2025 01:55
URI:	http://repository.its.ac.id/id/eprint/119560

Actions (login required)

View Item