Segmentasi Gumpalan Darah Vena Pada Citra Ultrasound Menggunakan U-Net

Ramadhani, Ahmad (2024) Segmentasi Gumpalan Darah Vena Pada Citra Ultrasound Menggunakan U-Net. Masters thesis, Institut Teknologi Sepuluh Nopember.

Text 6022221001-Master_Thesis.pdf - Accepted Version Restricted to Repository staff only until 1 April 2026. Download (5MB) | Request a copy

Abstract

Deep Vein Thrombosis (DVT) merupakan sebuah penyakit yang diakibatkan adanya pembentukan thrombus pada pembuluh darah vena dalam. Thrombus ini dapat mengganggu aliran darah normal dan menyebabkan masalah serius apabila tidak diobati. Dataset yang digunakan dalam penelitian ini berupa citra 2D ultrasound thrombus 5 pasien penderita DVT dan citra thrombus dan pembuluh darah phantom balon panjang. Diagnosis thrombus apabila dilakukan secara manual memerlukan waktu yang tidak sebentar serta analisis akurasi pembacaan citra thrombus bergantung pada dokter spesialis. Oleh karena itu, diperlukan adanya diagnosis thrombus untuk penderita DVT secara otomatis guna mempersingkat waktu serta meningkatkan performa analisis akurasi pembacaan citra thrombus. Penelitian ini mengusulkan segmentasi 2D dan 3D thrombus pada citra ultrasound phantom balon panjang menggunakan model segmentasi U-Net. Penelitian ini berhasil melakukan segmentasi gumpalan darah vena pada citra ultrasound menggunakan U-Net 3D. Berdasarkan hasil segmentasimodel segmentasi U-Net 3D mendapat nilai accuracy sebesar 99,1078% dan nilai loss sebesar 0,0208. Berdasarkan perhitungan evaluasi metrik untuk perhitungan antara hasil citra prediksi dan groundtruth dengan menggunakan IoU, dice coefficient, dan hausdorff distance, citra 3D ultrasound thrombus dan pembuluh darah mendapat nilai mean IoU sebesar 0,8105, mean dice coefficient sebesar 0,8953, dan mean hausdorff distance sebesar 3,25. Pada segmentasi 2D citra ultrasound thrombus dan pembuluh darah phantom balon panjang, penggunaan encoder pre-trained VGG16 pada model U-Net 2D dapat meningkatkan kinerja model untuk segmentasi area thrombus. Penerapan peningkatan kualitas citra dengan filter gaussian dan filter median memberikan pengaruh dalam
peningkatan performa segmentasi 2D.
==================================================================================================================================
Deep Vein Thrombosis (DVT) is a disease that occurs when a thrombus forms within the deep veins. This thrombus can disrupt normal blood flow and lead to severe issues if left untreated. The dataset used in this research consists of 2D ultrasound images of thrombus from 5 patients with DVT. Manual thrombus diagnosis requires a considerable amount of time, and the accuracy of thrombus image analysis relies on specialized doctors. Hence, an automatic thrombus diagnosis is needed for DVT patients to shorten the time and enhance the accuracy of thrombus image analysis. This study proposes 2D and 3D thrombus segmentation on long balloon phantom ultrasound images using the U-Net segmentation model. This research successfully performed venous blood clot segmentation in ultrasound images using 3D U-Net. Based on the segmentation results, the 3D U-Net segmentation model achieved an accuracy of 99.1078% and a loss value of 0.0208. Based on the evaluation metric calculations for the comparison between predicted image results and ground truth using IoU, dice coefficient, and Hausdorff distance, the 3D ultrasound images of thrombus and blood vessels obtained a mean IoU value of 0.8105, mean dice coefficient of 0.8953, and mean Hausdorff distance of 3.25. In the 2D segmentation of long balloon phantom ultrasound images of thrombus and blood vessels, the use of a pre-trained VGG16 encoder in the 2D U-Net model was able to enhance the model’s performance for thrombus area segmentation. The application of image quality improvements with Gaussian and median filters had an effect on improving the performance of 2D segmentation.

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	Deep Vein Thrombosis, Citra Ultrasound, Segmentasi, Pre-trained VGG16 dan UNet, Filter Denoising, Ultrasound Image, Segmentation, Denoising Filter
Subjects:	R Medicine > R Medicine (General) > R858 Deep Learning R Medicine > RC Internal medicine > RC691 Blood-vessels--Diseases. R Medicine > RC Internal medicine > RC78.7.U4 Ultrasonic imaging.
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Electrical Engineering > 20101-(S2) Master Thesis
Depositing User:	Ahmad Ramadhani
Date Deposited:	29 Jan 2024 01:36
Last Modified:	29 Jan 2024 01:36
URI:	http://repository.its.ac.id/id/eprint/105680

Actions (login required)

View Item