Faiz, Muhammad Abdurrahman (2026) Rekonstruksi 3D Pembuluh Darah Untuk Visualisasi Aneurisma Menggunakan Pipeline Segmentasi 3D U-Net Dua Jalur pada Citra TOF-MRA. Other thesis, Institut Teknologi Sepuluh Nopember.
|
Text
5025221058-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (7MB) |
Abstract
Model tiga dimensi (3D) yang akurat dari aneurisma otak dan pembuluh darah di sekitarnya sangat penting bagi dokter untuk menilai kondisi pasien serta merencanakan tindakan pembedahan. Metode segmentasi manual konvensional pada citra Time-of-Flight Magnetic Resonance Angiography (TOF-MRA) umumnya memerlukan waktu yang sangat lama, bergantung pada penilaian subjektif operator, dan kurang mampu menangani kompleksitas morfologi pada citra TOF-MRA. Meskipun metode berbasis deep learning untuk segmentasi otomatis telah banyak dikembangkan, sebagian besar masih menggunakan satu model tunggal untuk seluruh struktur dan belum mampu secara memadai membedakan karakteristik morfologis antara aneurisma dan pembuluh darah normal. Untuk mengatasi permasalahan tersebut, penelitian ini mengusulkan metode segmentasi dua jalur paralel (dual-path) yang memanfaatkan dua model 3D U-Net terpisah. Model pertama telah dipra-latih untuk mendeteksi aneurisma, sedangkan model kedua secara khusus dirancang untuk mempertahankan struktur pembuluh darah yang rinci dan saling terhubung. Hasil segmentasi dari kedua model kemudian digabungkan menggunakan metode fusi yang mempertimbangkan topologi pembuluh darah, diikuti dengan tahap pascapemrosesan untuk lebih menyempurnakan hasil segmentasi. Metode yang diusulkan mencapai nilai Dice Similarity Coefficient sebesar 0,9118, yang menunjukkan peningkatan sebesar 5,85% dibandingkan dengan penggunaan 3D U-Net konvensional. Selain itu, akurasi batas pada hasil rekonstruksi mengalami peningkatan yang signifikan, ditunjukkan oleh penurunan Average Hausdorff Distance sebesar 24,16%, dari 0,4369 mm menjadi 0,3373 mm. Hasil penelitian ini menunjukkan bahwa pemodelan yang mempertimbangkan karakteristik morfologis spesifik dari setiap struktur dapat menghasilkan rekonstruksi tiga dimensi pembuluh darah yang lebih akurat dan andal.
===============================================================================================================================
Accurate three-dimensional (3D) models of brain aneurysms and the blood vessels around them are vitally important for doctors to assess the condition and plan surgery.
Standard manual tracing of these features on Time-of-Flight Magnetic Resonance Angiography (TOF-MRA) images is usually very time-consuming, relies on the judgement of
the person doing it, and does not work well with the complex shapes of TOF-MRA. Deep learning programs that automatically do this are common, but most use a single program for all structures and do not sufficiently recognize the differences in form between an aneurysm and a normal blood vessel. To solve this, our research introduces a method with two parallelroutes, using two separate 3D U-Net programs. One has been trained beforehand to locate aneurysms, and the other is specifically for preserving the detailed, connected form of theb lood vessels. By combining the results of these two programs with a method of merging that understands the structure of the vessels and then improving the results afterwards, Dice Similarity Coefficient of 0.9118 was achieved. This is 5.85\% better than using a typical 3D U-Net. Importantly, the accuracy of the edges of the reconstructed images was considerably improved, shown by a 24.16% reduction in the Average Hausdorff Distance, from 0.4369 mm to 0.3373 mm. These results show that teaching programs to understand the specific shapes of different areas creates more accurate and dependable 3D images of blood vessels.
| Item Type: | Thesis (Other) |
|---|---|
| Uncontrolled Keywords: | TOF-MRA, Segmentasi, Deep Learning, 3D Reconstruction, Aneurysm, 3D Reconstruction, Aneurysm, Deep Learning, Segmentation, TOF-MRA |
| Subjects: | R Medicine > R Medicine (General) > R858 Deep Learning T Technology > T Technology (General) > T385 Visualization--Technique T Technology > T Technology (General) > T57.5 Data Processing T Technology > T Technology (General) > T58.64 Information resources management |
| Divisions: | Faculty of Information Technology > Informatics Engineering > 55201-(S1) Undergraduate Thesis |
| Depositing User: | Muhammad Abdurrahman Faiz |
| Date Deposited: | 30 Jul 2026 01:23 |
| Last Modified: | 30 Jul 2026 01:23 |
| URI: | http://repository.its.ac.id/id/eprint/138884 |
Actions (login required)
![]() |
View Item |
