Klasifikasi dan Segmentasi Lesi Covid-19 Pada Citra CT Scan Paru-Paru Menggunakan Multi-Task Deep Learning

Irsyad, Akhmad (2024) Klasifikasi dan Segmentasi Lesi Covid-19 Pada Citra CT Scan Paru-Paru Menggunakan Multi-Task Deep Learning. Doctoral thesis, Institut Teknologi Sepuluh Nopember.

Text 7025201009-Dissertation.pdf - Accepted Version Restricted to Repository staff only until 1 October 2026. Download (3MB) | Request a copy

Abstract

Coronavirus disease 2019 (COVID-19) adalah penyakit disebabkan oleh virus Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2). Penyakit COVID-19 menyebar dengan cepat dan dapat menyebabkan kematian. Selama fase awal penyakit ini ditandai dengan demam, batuk, dan kelelahan. Penelitian menunjukkan bahwa virus COVID-19 menyebar dari orang ke orang. Orang yang terinfeksi memiliki masalah pernapasan serius dan perlu dirawat di unit perawatan intensif. Para pasien memiliki kondisi abnormal pada citra paru-paru hasil Computed Tomography (CT). CT scan orang yang terinfeksi menunjukkan bahwa penyakit COVID-19 memiliki karakteristik sendiri. Karena itu, para ahli klinis membutuhkan citra CT paru untuk mendiagnosis COVID-19 pada fase awal. Multi-task deep learning adalah bagian dari deep learning yang bertujuan untuk menyelesaikan beberapa tugas berbeda secara bersamaan. Pada penelitian ini dikembangkan model untuk klasifikasi dan segmentasi lesi COVID-19 berdasarkan CT paru-paru menggunakan multi-task deep learning. Proses segmentasi menggunakan modifikasi pada block decoding Unet dengan menggunakan aktivasi swish untuk meningkatkan kinerja dari model. Menggunakan multi-task deep learning diharapkan dapat mengurangi kemungkinan terjadinya overfitting. Dari pelatihan yang dilakukan diperoleh bahwa dengan menggunakan arsitektur Resnet50 MTL + SwishUnet dengan attention gate diperoleh kinerja yang lebih baik apabila dibandingkan dengan arsitektur single task learning baik itu untuk segmentasi dan klasifikasi. MTL Resnet50+swishUnet dengan attention gate memiliki kinerja terbaik dimana untuk segmentasi diperoleh akurasi 98,75%, sensitivity 84,58%, dan specificity 98,82%, untuk klasifikasi diperoleh kinerja akurasi 94,56%, precision 92,15%, recall 97,91%, dan F-measure 94,94%.
=================================================================================================================================
Coronavirus disease 2019 (COVID-19) is a disease caused by Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2). The COVID-19 spreads rapidly and can cause death. The early phase of the disease is characterized by fever, cough, and fatigue. The researchers found the COVID-19 virus spreads from person to person. The infected person has severe breathing problems and needs to be admitted to the intensive care unit. The patients had abnormal conditions on computed tomography (CT) lung images. A CT scan of an infected person shows that the COVID-19 disease has its own characteristics. Therefore, clinical experts need CT images of the lungs to diagnose COVID-19 early.
Multi-task deep learning is a part of deep learning that aims to complete several different tasks simultaneously. This research developed a model for the classification and segmentation of COVID-19 lesions based on lung CT using multi-task deep learning. The segmentation process uses modifications to the Unet decoding block using swish activation to improve the model's performance. Using multi-task deep learning is expected to reduce the possibility of overfitting. From the training carried out, it was found that using the Resnet50 MTL + SwishUnet architecture with attention gate, better performance was obtained compared to the single-task learning architecture for both segmentation and classification. MTL Resnet50+swishUnet with attention gate has the best performance where for segmentation the accuracy is 98.75%, sensitivity 84.58%, and specificity 98.82%, for classification the performance is 94.56% accuracy, precision 92.15%, recall 97.91%, and F-measure 94.94%.

Item Type:	Thesis (Doctoral)
Uncontrolled Keywords:	CT scan paru-paru, COVID-19, klasifikasi, segmentasi, Multi-task deep learning, classification, segmentation
Subjects:	Q Science > QA Mathematics > QA336 Artificial Intelligence Q Science > QA Mathematics > QA76.87 Neural networks (Computer Science)
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55001-(S3) PhD Thesis (Comp Science)
Depositing User:	Akhmad Irsyad
Date Deposited:	28 Aug 2024 04:49
Last Modified:	28 Aug 2024 04:49
URI:	http://repository.its.ac.id/id/eprint/114987

Actions (login required)

View Item