Arsitektur Adaptive EfficientNet-DeepLabV3+ untuk Segmentasi Dinding Ventrikel Kiri pada Citra Ultrasound Jantung

Sukma, Yoga Aji (2024) Arsitektur Adaptive EfficientNet-DeepLabV3+ untuk Segmentasi Dinding Ventrikel Kiri pada Citra Ultrasound Jantung. Masters thesis, Institut Teknologi Sepuluh Nopember.

Text 6003212001-Master_Thesis.pdf - Accepted Version Restricted to Repository staff only until 1 July 2026. Download (10MB) | Request a copy

Abstract

Elektrokardiografi (EKG), umumnya digunakan sebagai prosedur tradisional untuk melakukan diagnosis pada pasien penyakit jantung. Mengamati struktur dan pergerakan dinding ventrikel kiri dengan citra ultrasound yang dihasilkan dari prosedur EKG dapat digunakan untuk membantu mendiagnosa penyakit jantung, salah satunya Myocardial Infarction (MI). Namun, citra ultrasound memiliki banyak noise dan kualitas citra yang tidak bagus, sehingga metode segmentasi dapat diadaptasi untuk memperjelas area yang diamati, seperti dinding ventrikel kiri. Menjawab tantangan tersebut, penelitian ini melakukan segmentasi dinding ventrikel kiri dengan arsitektur deep learning yang secara adaptif menyesuaikan lebar dan kedalaman jaringan berdasarkan resolusi input, yang disebut Adaptive EfficientNet-DeepLabV3+. Arsitektur ini dihasilkan dari mengganti encoder DeepLabV3+, yaitu ResNet50, dengan arsitektur EfficientNet yang telah dimodifikasi dengan blok adaptif. Selanjutnya, pengujian kinerja arsitektur Adaptive EfficientNet-DeepLabV3+ dibandingkan dengan model DeepLabV3+ dengan encoder ResNet50. Seluruh prosedur pengujian dan pembuatan model dilakukan dengan memberikan variasi pada resolusi input menggunakan dataset publik yang dipublikasi oleh Hamad Medical Corporation, Qatar University, dan Tampere University (HMC-QU). Model Adaptive EfficientNet-DeepLabV3+ pada scenario 224 pixel memiliki kinerja lebih baik berdasarkan nilai pengukuran metrik evaluasi F1-Score yang lebih tinggi, waktu proses lebih cepat, dan loss yang lebih rendah. Model Adaptive EfficientNet-DeepLabV3+ yang diperoleh digunakan untuk membuat dashboard yang dapat membantu visualisasi hasil segmentasi dinding ventrikel kiri.
===================================================================================================================================
Electrocardiography (EKG) is commonly used as a traditional procedure for diagnosing heart diseases in patients. Observing the structure and movement of the left ventricle wall with ultrasound images generated from the EKG procedure can aid in diagnosing heart diseases, including myocardial infarction (MI). However, ultrasound images often contain a considerable amount of noise and exhibit poor quality, prompting the adaptation of segmentation methods to clarify the observed areas, such as the left ventricle wall. In response to this challenge, this research researches the left ventricle wall segmentation using a deep learning architecture that adaptively adjusts the width and depth of the network based on input resolution, referred to as Adaptive EfficientNet-DeepLabV3+. This architecture is derived from replacing the DeepLabV3+ encoder, namely ResNet50, with an EfficientNet architecture modified with adaptive blocks. Subsequently, the performance of the Adaptive EfficientNet-DeepLabV3+ architecture is compared with the DeepLabV3+ model using the ResNet50 encoder. The entire testing and model creation procedure is done by varying input resolutions using public datasets published by Hamad Medical Corporation, Qatar University, and Tampere University (HMC-QU). The Adaptive EfficientNet-DeepLabV3+ model in the 224-pixel scenario performs better based on higher F1-Score evaluation metric measurement values, faster processing time, and lower loss. The Adaptive EfficientNet-DeepLabV3+ model obtained was used to create a dashboard that can help visualize the results of left ventricular wall segmentation.

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	Adaptive EfficientNet-DeepLabV3+, Citra Ultrasound, Dinding Ventrikal Kiri, Segmentasi; Adaptive EfficienNet-DeeplabV3+, Left Ventricular Wall, Segmen-tation, Ultrasound Image
Subjects:	Q Science > Q Science (General) > Q325.78 Back propagation Q Science > QA Mathematics > QA336 Artificial Intelligence Q Science > QA Mathematics > QA76.87 Neural networks (Computer Science) Q Science > QA Mathematics > QA76.9.U83 Graphical user interfaces. User interfaces (Computer systems)--Design.
Divisions:	Faculty of Science and Data Analytics (SCIENTICS) > Statistics > 49101-(S2) Master Thesis
Depositing User:	Yoga Aji Sukma
Date Deposited:	20 Feb 2024 01:10
Last Modified:	20 Feb 2024 01:10
URI:	http://repository.its.ac.id/id/eprint/107604

Actions (login required)

View Item