ResNet-UNet dengan Transfer Learning untuk Segmentasi Ruang Ventrikel pada Video Ultrasound Jantung Anak

Nuraini, Ulfa Siti (2021) ResNet-UNet dengan Transfer Learning untuk Segmentasi Ruang Ventrikel pada Video Ultrasound Jantung Anak. Masters thesis, Institut Teknologi Sepuluh Nopember.

[img] Text
06211950010016-Master_Thesis.pdf - Accepted Version
Restricted to Repository staff only until 1 October 2023.

Download (5MB) | Request a copy

Abstract

Penyakit Jantung Bawaan (PJB) merupakan kasus yang sering terjadi pada anak-anak. PJB ini terjadi pada 1% bayi baru lahir di Indonesia. Salah satu jenis PJB yang sering terjadi yaitu VSD (Ventricular Septal Defect). Penyakit ini terjadi pada bayi baru lahir dengan adanya lubang antara ruang ventrikel kanan dan kiri pada jantung anak. Dalam pemeriksaan lanjut, dokter perlu melihat video ultrasound dari alat ekokardiografi dengan cara mendeteksi ruang ventrikel. Dalam penelitian ini, dilakukan segmentasi ruang ventrikel kanan dan kiri pada setiap frame video ultrasound. U-Net terdiri atas encoder dan decoder, dimana terdapat skip connection antar keduanya. Penelitian ini bertujuan untuk memperoleh pengembangan arsitektur baru bernama ResNet-UNet dengan mengganti encoder U-Net dengan ResNet yang dilengkapi transfer learning. ResNet yang digunakan yaitu ResNet18, ResNet34, ResNet50, ResNet101, dan ResNet152. Proses segmentasi ini dilakukan pada 15 video ultrasound pasien VSD di RSUD Dr. Soetomo, Surabaya. Model ResNet-UNet dengan transfer learning mendapatkan hasil yang lebih baik daripada U-Net. Metode yang memiliki kinerja klasifikasi yang konsisten yaitu ResNet101-UNet dengan transfer learning dimana menghasilkan rata-rata nilai Jaccard Index pada data validation mencapai nilai 82,6%. Hasil segmentasi pada data testing mendapatkan ROI yang cukup mendekati gambar ground truth dengan nilai Jaccard Index pada data testing sebesar 70,9%. Dari model terbaik itu dibuat dashboard untuk menampilkan hasil segmentasi. Sehingga penelitian ini akan dapat membantu paramedis dengan lebih cepat mengambil tindakan medis berdasarkan pada hasil segmentasi dan dashboard yang telah dihasilkan. ===================================================================================================== Congenital Heart Disease (CHD) is a case that often occurs in children. CHD occurs in 1% of newborns in Indonesia. One type of CHD that often occurs is VSD (Ventricular Septal Defect). This disease occurs in newborns with a hole between the right and left ventricular spaces in the child's heart. In further examination, the doctor needs to see the ultrasound video from the echocardiography device by detecting the ventricular space. Therefore, in this study, we performed segmentation of the right and left ventricular spaces on each ultrasound video frame. U-Net is one of the most reliable segmentation methods in the medical sector. U-Net consists of an encoder and a decoder, where there are skip connections between them. This study aims to obtain the development of a new architecture called ResNet-UNet by replacing the U-Net encoder with ResNet equipped with transfer learning. ResNet used are ResNet18, ResNet34, ResNet50, ResNet101, and ResNet152. This segmentation process was carried out on 15 ultrasound videos of VSD patients at Dr. Soetomo Hospital, Surabaya. ResNet-UNet model with transfer learning got better results than U-Net. The method that has a consistent classification performance is ResNet101-UNet with transfer learning which produces an average Jaccard Index value on data validation reaching a value of 82.6%. The results of segmentation on testing data get a ROI image that is quite close to the ground truth image with the average of Jaccard Index value on the testing data of 70.9%. The dashboard has been created to display the segmentation results based on the best model. So, this research will be able to help paramedics more quickly take medical action based on the results of segmentation and dashboards that have been obtained.

Item Type: Thesis (Masters)
Uncontrolled Keywords: ResNet, Segmentasi, U-Net, Video Ultrasound, Ventrikel ResNet, Segmentation, Ultrasound Video, U-Net, Ventricle
Subjects: Q Science > Q Science (General) > Q325.5 Machine learning.
Q Science > QA Mathematics > QA336 Artificial Intelligence
Q Science > QA Mathematics > QA76.87 Neural networks (Computer Science)
Divisions: Faculty of Science and Data Analytics (SCIENTICS) > Statistics > 49101-(S2) Master Thesis
Depositing User: ULFA SITI NURAINI
Date Deposited: 09 Sep 2021 03:30
Last Modified: 09 Sep 2021 03:30
URI: https://repository.its.ac.id/id/eprint/91905

Actions (login required)

View Item View Item