TEKNIK SLICE-BASED DAN ARSITEKTUR 2D-DENSE-MOBILENET PADA CITRA 3D CHEST CT-SCAN UNTUK KLASIFIKASI COVID-19

SETYAWATI, INDA and Desiani, Anita and Andriani, Yuli (2023) TEKNIK SLICE-BASED DAN ARSITEKTUR 2D-DENSE-MOBILENET PADA CITRA 3D CHEST CT-SCAN UNTUK KLASIFIKASI COVID-19. Undergraduate thesis, Sriwijaya University.

	Text RAMA_44201_08011381924059.pdf - Accepted Version Restricted to Repository staff only Available under License Creative Commons Public Domain Dedication. Download (1MB) | Request a copy
	Text RAMA_44201_08011381924059_TURNITIN.pdf - Accepted Version Restricted to Repository staff only Available under License Creative Commons Public Domain Dedication. Download (7MB) | Request a copy
	Text RAMAA_44201_08011381924059_0011127702_0002077202_01_front_ref.pdf - Accepted Version Available under License Creative Commons Public Domain Dedication. Download (449kB)
	Text RAMAA_44201_08011381924059_0011127702_0002077202_02.pdf - Accepted Version Restricted to Repository staff only Available under License Creative Commons Public Domain Dedication. Download (768kB) | Request a copy
	Text RAMAA_44201_08011381924059_0011127702_0002077202_03.pdf - Accepted Version Restricted to Repository staff only Available under License Creative Commons Public Domain Dedication. Download (175kB) | Request a copy
	Text RAMAA_44201_08011381924059_0011127702_0002077202_04.pdf - Accepted Version Restricted to Repository staff only Available under License Creative Commons Public Domain Dedication. Download (1MB) | Request a copy
	Text RAMAA_44201_08011381924059_0011127702_0002077202_05.pdf - Accepted Version Restricted to Repository staff only Available under License Creative Commons Public Domain Dedication. Download (141kB) | Request a copy
	Text RAMAA_44201_08011381924059_0011127702_0002077202_06_ref.pdf - Bibliography Restricted to Repository staff only Available under License Creative Commons Public Domain Dedication. Download (192kB) | Request a copy

Abstract

Convolutional neural network (CNN) is a deep learning method that has many layers and requires large data. However, image data sizes that are too large can affect computational performance due to large memory usage. Limited memory can cause CNN not to work optimally. The solution to overcome the large number of data sizes and multiply the data in 3D images to meet the needs of CNN training, namely, images can be sliced into 2D images using slice-based techniques to improve the training function on CNN. One of the CNN architectures is MobileNet which has light computation and high speed because it produces small parameters, but small parameters are not enough to study features in large data. Lack of parameters in training can affect classification performance to be not optimal. One way to increase parameters is by using dense-blocks. Dense-block is a block layer on CNN which has a feed forward layer that is connected to each previous layer to another layer. Dense-block additions to the MobileNet architecture to increase parameters and capture more complex features. This study applies slice-based techniques and 2D Dense-MobileNet architecture to meet the data requirements on CNN and obtain a CNN architecture that is able to improve performance on the MobileNet architecture. The stages carried out in this classification process are preprocessing, data augmentation, training, and testing. The results of the study with the 3D CT-Scan chest dataset obtained an accuracy value of 94.41%, a sensitivity of 88.45%, a specificity of 96.07%, an f1-score of 88.21%, and Cohen's kappa of 85.44%. Based on these results, it shows that the 2D-Dense-MobileNet architecture is capable of performing classification tasks to determine the level of COVID-19 infection from the image data used.

Item Type:	Thesis (Undergraduate)
Uncontrolled Keywords:	Klasifikasi, CT-Scan,teknik slice-base, MobileNet, Dense-block
Subjects:	Q Science > QA Mathematics > QA1-43 General
Divisions:	08-Faculty of Mathematics and Natural Science > 44201-Mathematics (S1)
Depositing User:	INDA SETYAWATI
Date Deposited:	30 Oct 2023 07:15
Last Modified:	30 Oct 2023 07:15
URI:	http://repository.unsri.ac.id/id/eprint/130146

Actions (login required)

View Item