Pengembangan Kerangka Kerja Segmentasi 2.5D Inferior Alveolar Nerve Pada Citra CBCT Menggunakan Ensemble Attention U-Net

Tama, Arsy Bilahil Tama (2026) Pengembangan Kerangka Kerja Segmentasi 2.5D Inferior Alveolar Nerve Pada Citra CBCT Menggunakan Ensemble Attention U-Net. Masters thesis, Institut Teknologi Sepuluh Nopember.

Text 6025221029_Master_Thesis.pdf Restricted to Repository staff only Download (3MB) | Request a copy

Abstract

Segmentasi otomatis Inferior Alveolar Nerve (IAN) pada citra Cone Beam Computed Tomography (CBCT) sangat dibutuhkan untuk perencanaan implan gigi, karena anotasi manual memakan waktu dan menghasilkan variasi antar-ahli yang tinggi. Tesis ini mengusulkan kerangka kerja segmentasi IAN berbasis citra 2.5D yang menggabungkan klasifikasi dan segmentasi. Volume CBCT didekomposisi menjadi tiga proyeksi ortogonal (aksial, sagital, koronal). Pada setiap proyeksi, klasifikasi Ensemble digunakan sebagai Fully connected layer untuk menyaring irisan yang mengandung IAN, lalu segmentasi dilakukan menggunakan Attention U-Net. Hasil tiap proyeksi direkonstruksi menjadi volume 3D dan digabungkan melalui Soft weighted fusion. Pada sembilan pasien uji, metode yang diusulkan mencapai Dice Score 0,798 ± 0,061 dan akurasi klasifikasi Fully connected layer 94,8–96,4%, berada pada rentang atas metode referensi berbasis 3D pada dataset ToothFairy. Pendekatan 2.5D yang diusulkan juga terbukti lebih efisien secara memori GPU dibandingkan segmentasi 3D volumetric penuh, sehingga lebih realistis untuk diterapkan pada infrastruktur klinis dengan resource terbatas.
==================================================================================================================================
Automated Segmentation of the Inferior Alveolar Nerve (IAN) on Cone Beam Computed Tomography (CBCT) is essential for dental implant planning, as manual annotation is time-consuming and shows high inter-observer variability. This thesis proposes a 2.5D-based IAN segmentation framework that combines classification and segmentation. The CBCT volume is decomposed into three orthogonal projections (axial, sagittal, and coronal). For each projection, an Ensemble classifier serves as the Fully connected layer stage to filter IAN-containing slices, after which segmentation is performed using an Attention U-Net. The results from each projection are reconstructed into a 3D volume and combined via a Soft weighted fusion strategy. On nine test patients, the proposed method achieves a Dice Score of 0.798 ± 0.061 and a Fully connected layer classification accuracy of 94.8–96.4%, within the upper range of 3D reference methods on the ToothFairy dataset. The proposed 2.5D approach is also demonstrated to be more GPU memory-efficient than full 3D volumetric segmentation, making it more feasible for deployment on clinical infrastructure with limited resources.

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	Deep learning, Inferior alveolar Nerve, Segmentasi, CBCT, Attention,U-Net Deep learning, Inferior Alveolar Nerve, Segmentation, CBCT, Attention U-Net
Subjects:	R Medicine > R Medicine (General) > R858 Deep Learning
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55101-(S2) Master Thesis
Depositing User:	Arsy Bilahil Tama
Date Deposited:	30 Jul 2026 03:51
Last Modified:	30 Jul 2026 03:51
URI:	http://repository.its.ac.id/id/eprint/140222

Actions (login required)

View Item