Darmawan, Achmad Hanan (2022) Segmentasi Ground Glass Opacity Pada Citra X-Ray Paru COVID-19 Menggunakan Jaringan Adversarial Berbasis U-NET. Other thesis, Institut Teknologi Sepuluh Nopember.
Preview |
Text
07311740000035-Undergraduate_Thesis.pdf - Accepted Version Download (4MB) | Preview |
Abstract
COVID-19 merupakan jenis virus yang menyebabkan wabah di seluruh dunia hari ini. Virus ini menyerang saluran pernapasan manusia dan sangat menular. Pemeriksaan radiografi seperti X-Ray dan computed tomography (CT) telah digunakan untuk mengidentifikasi pola morfologis lesi paru-paru terkait COVID-19. Citra X-Ray paru-paru pasien COVID-19 menyimpan informasi terkait kondisi klinis pasien, dimana terdapat objek patologi yang dapat dideteksi pada pasien COVID-19. Pendeteksian objek patologi dilakukan secara manual oleh radiologist sehingga pengaplikasiannya bergantung pada pengamatan radiologist dan ketersediaan radiologist. Oleh karena itu, saat ini diperlukan suatu penelitian mengenai metode segmentasi X-Ray paru-paru otomatis berbasis deep learning dengan menggunakan Convolutional Neural Network untuk mensegmentasi objek patologi ground glass opacity. Dataset yang digunakan adalah Dataset X-Ray paru-paru COVID-19 berbasis open-source yang tersedia di Github. Tahapan pengolahan citra X-Ray paru-paru COVID-19 diawali dengan tahapan pre-processing untuk menormalisasi piksel, lalu menyamaratakan grey-level citra, dan me-resizing ukuran dataset citra. Tahap selanjutnya adalah segmentasi ground glass opacity dengan menggunakan CNN arsitektur jaringan adversarial. Hasil pembelajaran model segmentasi ground glass opacity ini memiliki performa Dice Coefficient (DSC) mencapai 88.6% dan performa pengujian model dengan hasil DSC 50.1%, recall 0.474, precission 0.532, true positive 0.474, true negative 0.846.
====================================================================================================
COVID-19 is a virus that causes outbreaks around the world today. The virus attacks the human respiratory tract and is highly contagious. Radiographic examinations such as X-Ray and computed tomography (CT) have been used to identify morphological patterns of COVID-19 related lung lesions. Lungs X-Ray images of COVID-19 hold information related to the clinical condition of patient, where there are pathology objects that can be detected in COVID-19 patients. Detection of pathological objects is done manually by radiologists so that their application relies on radiologist observation and is limited to radiological availability. Therefore, a study is currently needed on the automatic lung X-Ray segmentation method based on deep learning by using the Convolutional Neural Network to segment the object of ground glass opacity pathology. The dataset used is an open-source COVID-19 lung X-Ray Dataset available on Github. The processing stage of COVID-19 lung X-Ray imagery begins with a pre-processing stage to normalize pixels, then generalize grey-level imagery, and resizing the size of the image dataset. The next stage is segmentation of ground glass opacity using CNN's adversarial network architecture. The results of this ground glass opacity segmentation model have a Dice Coefficient (DSC) performance of up to 88.6% and a model testing performance with DSC results of 50.1%, recall 0.474, precision 0.532, true positive 0.474, true negative 0.846.
Actions (login required)
 |
View Item |