Deteksi Ventricular Septal Defect Berdasarkan Video Ultrasound Jantung Anak pada Parasternal dan Apical View dengan Faster Region-based Convolutional Neural Network

Nirmalasari, Nur Indah (2021) Deteksi Ventricular Septal Defect Berdasarkan Video Ultrasound Jantung Anak pada Parasternal dan Apical View dengan Faster Region-based Convolutional Neural Network. Masters thesis, Institut Teknologi Sepuluh Nopember.

[img] Text
06211950010015-Master_Thesis.pdf - Accepted Version
Restricted to Repository staff only until 1 October 2023.

Download (3MB) | Request a copy

Abstract

Terdapat 9 dari 1000 bayi lahir di Indonesia yang menderita penyakit jantung bawaan (PJB). Ventricular septal defect (VSD) merupakan jenis PJB yang paling sering ditemukan pada anak-anak. Anak yang hidup dengan PJB beresiko mengalami keterlambatan dan gangguan perkembangan. Salah satu pemeriksaan PJB adalah ekokardiografi, yaitu pemeriksaan medis dengan teknologi ultrasound yang dapat memberikan citra jantung anak secara real-time. Tujuan dari penelitian ini adalah membantu visualisasi lokasi VSD pada jantung anak berdasarkan video ultrasound jantung dari view parasternal dan apical dengan menggunakan Faster R-CNN. Model object detection ini mampu menghasilkan akurasi yang baik, terutama dengan pemilihan feature extractor yang tepat. Pada penelitian ini pemodelan dilakukan menggunakan beberapa feature extractor yaitu ResNet, FPN dan Inception-Resnet V2. Data yang digunakan dalam penelitian ini diambil dari RSUD Dr. Soetomo. Model dilatih dengan frame yang diekstraksi dari video ultrasound jantung dan dipilih frame yang menampakkan VSD. Model-model dengan feature extractor berbeda yang telah dilatih dibandingkan menggunakan ukuran evaluasi average precision (AP). Diperoleh bahwa Faster R-CNN based Inception ResNet V2 menghasilkan AP validasi dan testing yang paling baik padaview apical. Sedangkan untuk data view parasternal long axis, model Faster R�CNN based ResNet50 memberikan nilai AP test yang cukup baik untuk dataset validasi maupun testing. ==================================================================================================== There are 9 out of 1000 babies born in Indonesia have congenital heart disease (CHD). Ventricular septal defect (VSD) is the most common type of CHD in children. Children that living with CHD have a high risk in developmental delays and disorders. One of the examinations for CHD is echocardiography, which is a medical examination with ultrasound technology that can provide real-time images of the child's heart. The purpose of this study was to help visualize the location of the VSD in the child's heart based on cardiac ultrasound video from parasternal and apical views using Faster R-CNN. This object detection model can produce good accuracy, especially with the selection of the right feature extractor. In this study, models were trained using several feature extractors, namely ResNet, FPN, and Inception-Resnet V2. The data used in this study were collected from RSUD Dr. Soetomo. Frames were extracted from ultrasound videos, then we selected frames that have visible VSD for model training. Models with different feature extractors that have been trained are compared using the average precision (AP). As a result, Faster R-CNN-based Inception ResNet V2 produced high AP in validation and testing dataset for apical view data. For the long axis parasternal view data, the Faster R-CNN-based ResNet50 model can provide good AP in validation and testing datasets.

Item Type: Thesis (Masters)
Uncontrolled Keywords: Ekokardiografi, Faster R-CNN, PJB, Ultrasound, VSD, CHD, Echocardiography, Faster R-CNN, Ultrasound, VSD
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: Nur Indah Nirmalasari
Date Deposited: 10 Sep 2021 02:23
Last Modified: 10 Sep 2021 02:23
URI: https://repository.its.ac.id/id/eprint/91941

Actions (login required)

View Item View Item