Rekonstruksi Permukaan 3d Infeksi Covid-19 Pada Citra Paru-Paru Ct-Scan Menggunakan Alpha Shape Berbasis Segmentasi U-Net

Ferdinandus, F.X. (2023) Rekonstruksi Permukaan 3d Infeksi Covid-19 Pada Citra Paru-Paru Ct-Scan Menggunakan Alpha Shape Berbasis Segmentasi U-Net. Doctoral thesis, Institut Teknologi Sepuluh Nopember.

Text 7022202009-Dissertation.pdf - Accepted Version Restricted to Repository staff only until 1 October 2025. Download (13MB) | Request a copy

Abstract

Infeksi Covid-19 memaksa radiolog dan petugas medis untuk melakukan diagnosa dengan cepat terhadap pasien, hanya dalam beberapa hari saja infeksi virus Covid-19 telah mempengaruhi kinerja dari paru-paru untuk pernafasan. Ground Glass Opacity (GGO) berbentuk kabut tipis berwarna abu-abu sebagai akibat infeksi Covid-19 banyak dideteksi di dalam area paru-paru pasien. Di sisi lain penggunaan Deep Learning untuk melakukan segmentasi citra dalam beberapa tahun terakhir ini menjadi perhatian peneliti bahkan sebelum pandemi Covid-19 terjadi. Dalam penelitian ini diusulkan untuk mengetahui kondisi paru paru pasien Covid-19 melalui pendekatan Deep Learning dengan cara melakukan segmentasi terhadap paru-paru dan infeksi Covid-19 dalam area paru-paru menggunakan U-Net arsitektur. Hasil segmentasi akan digunakan untuk Visualisasi 3D dengan cara melakukan stack terhadap potongan citra hasil segmentasi, mendeteksi contour, membangun point cloud dan membuat rekonstruksi permukaan 3D Mesh melalui metode Alpha Shape. Langkah terakhir adalah membuat prediksi persentase volumetric infeksi terhadap paru-paru pasien.
Evaluasi hasil segmentasi dilakukan dengan menggunakan metrik IoU, Dice, Precision dan Accuracy, sedangkan evaluasi volumetric visualisasi 3D menggunakan metrik Relative Volume Difference (RVD) dan Volumetric Similarity (VS). Hasil evaluasi segmentasi menggunakan metrik Dice pada 3 set data pasien uji mendapatkan skor rata-rata 95% untuk paru-paru, dan mendapatkan skor rata-rata 75% untuk Infeksi Covid-19. Sedangkan hasil evaluasi prediksi persentase volume infeksi mempunyai selisih paling banyak 3% dari nilai ground-truth.
==================================================================================================================================
Covid-19 infection forces radiologists and medical workers to quickly diagnose patients, in just a few days the Covid-19 virus infection has affected the performance of the lungs for breathing. Ground Glass Opacity (GGO) in the form of a thin gray mist as a result of Covid-19 infection is mostly detected in the patient's lung area. On the other hand, the use of Deep Learning to perform image segmentation in recent years has been a concern of researchers even before the Covid-19 pandemic occurred. In this study it is proposed to determine the condition of the lungs of Covid-19 patients through a Deep Learning approach by segmenting the lungs and Covid-19 infection in the lung area using the U-Net architecture. The segmentation results will be used for 3D Visualization by stacking segmented image pieces, detecting contours, constructing point clouds and creating a 3D Mesh surface reconstruction using the Alpha Shape method. The final step is to make predictions of the volumetric percentage of infection in the patient's lungs. Evaluation of the segmentation results was carried out using the IoU, Dice, Precision and Accuracy metrics, while the 3D volumetric visualization evaluation used the Relative Volume Difference (RVD) and Volumetric Similarity (VS) metrics. The results of the segmentation evaluation using the Dice metric on 3 test patient data sets obtained an average score of 95% for the lungs, and an average score of 75% for Covid-19 infection. While the results of the evaluation of the prediction of the percentage of infection volume have a difference of at most 3% from the ground-truth value.

Item Type:	Thesis (Doctoral)
Uncontrolled Keywords:	Covid-19, Deep Learning, Lung segmentation, Semantic segmentation, U-Net
Subjects:	T Technology > T Technology (General) T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK7882.P3 Pattern recognition systems
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Electrical Engineering > 20001-(S3) PhD Thesis
Depositing User:	F.X. Ferdinandus
Date Deposited:	02 Aug 2023 03:20
Last Modified:	02 Aug 2023 03:20
URI:	http://repository.its.ac.id/id/eprint/100604

Actions (login required)

View Item