Identifikasi Penyakit Paru Obstruktif Kronik Menggunakan Graph Convolutional Network Pada Electronic Nose

Aulia, Dava (2024) Identifikasi Penyakit Paru Obstruktif Kronik Menggunakan Graph Convolutional Network Pada Electronic Nose. Masters thesis, Institut Teknologi Sepuluh Nopember.

Text 6025221056-Master_Thesis.pdf - Accepted Version Restricted to Repository staff only until 1 April 2026. Download (5MB) | Request a copy

Abstract

Saat ini, sistem electronic nose telah banyak digunakan sebagai metode non-invasif untuk mendiagnosis penyakit paru berdasarkan udara napas yang dihembuskan. Namun demikian, beberapa permasalahan masih perlu diselesaikan, antara lain jumlah dan jenis sensor gas yang optimal untuk penyakit paru obstruktif kronik (PPOK), penentuan fitur sensor gas yang sesuai untuk tujuan klasifikasi, dan algoritma klasifikasi yang dapat memberikan accuracy yang lebih tinggi. Oleh karena itu, penelitian ini mengembangkan sistem electronic nose, baik hardware dan software, untuk mengidentifikasi PPOK. Sistem electronic nose yang dikembangkan terdiri dari dua puluh sensor gas semikonduktor oksida logam (metal-oxide-semiconductor, MOS) untuk mendapatkan pola respons spesifik untuk setiap subyek, meliputi subyek sehat dan PPOK, melalui sampel udara napas yang dihembuskan. Sampel yang diperoleh untuk subyek PPOK sebesar 40 dan subyek sehat sebesar 30. Algoritma fast Fourier transform (FFT) diterapkan untuk mendapatkan elemen frekuensi pada respons sensor gas pada durasi tertentu. Teknik principal component analysis (PCA) digunakan untuk memvisualisasikan distribusi data dan bertindak sebagai pra-pemrosesan data. Algoritma firefly, yang disertai dengan algoritma support vector machine (SVM) sebagai fungsi fitness-nya, diimplementasikan untuk menentukan jumlah sensor gas yang optimal. Metode statistik Pearson correlation coefficient (PCC) diperkerjakan untuk memperoleh nilai koefisien korelasi antar fitur sensor gas dan dikuantifikasi untuk membentuk matriks adjacency dari algoritma graph convolutional network (GCN). Struktur multi-layer algoritma GCN dirancang untuk mencapai identifikasi PPOK. Hasil percobaan menunjukkan bahwa algoritma GCN yang dikombinasikan dengan teknik PCA pada dataset frekuensi dapat memisahkan kelompok subyek sehat dan PPOK secara lebih signifikan dengan nilai rata-rata accuracy sebesar 92,9%, precision sebesar 93,5%, recall atau sensitivity sebesar 92,1%, F1-score sebesar 92,6%, dan specificity sebesar 92,1%. Selain itu, algoritma firefly telah menghasilkan sebelas sensor gas yang memberikan pola respons sensor gas tertentu pada sampel udara napas yang dihembuskan dari subyek PPOK dan menghasilkan nilai rata-rata accuracy sebesar 88,6%, precision sebesar 90,3%, recall atau sensitivity sebesar 87,1%, F1-score sebesar 87,9%, dan specificity sebesar 87,1%.
=================================================================================================================================
Currently, electronic nose systems have been widely used as a non-invasive method for diagnosing lung diseases based on exhaled breath. However, several problems still need to be resolved, including the optimal number and type of gas sensors for chronic obstructive pulmonary disease (COPD), determining appropriate gas sensor features for classification purposes, and classification algorithms that can provide higher accuracy. Therefore, this research developed an electronic nose system, both hardware and software, to identify COPD. The developed electronic nose system comprises twenty metal-oxide-semiconductor (MOS) gas sensors to acquire specific response patterns for each subject, including healthy and COPD subjects, through exhaled breath samples. The samples obtained for COPD subjects were 40 and 30 for healthy subjects. The fast Fourier transform (FFT) algorithm is applied to receive the frequency element in the gas sensor response for a certain duration. The principal component analysis (PCA) technique is used to visualize data distribution and acts as data pre-processing. The firefly algorithm, accompanied by the support vector machine (SVM) algorithm as its fitness function, is implemented to determine the optimal number of gas sensors. The Pearson correlation coefficient (PCC) statistical method is employed to obtain correlation coefficient values between gas sensor features and is quantified to form an adjacency matrix from a graph convolutional network (GCN) algorithm. The multi-layer structure of the GCN algorithm is designed to achieve COPD identification. The experimental results show that the GCN algorithm combined with the PCA technique on the frequency dataset can separate groups of healthy and COPD subjects more significantly with an average accuracy value of 92.9%, precision of 93.5%, recall or sensitivity of 92.1%, F1-score of 92.6%, and specificity of 92.1%. In addition, the firefly algorithm has determined eleven gas sensors that provide certain gas sensor response patterns on exhaled breath samples from COPD subjects and produces an accuracy value of 88.6%, precision of 90.3%, recall or sensitivity of 87.1%, F1-score of 87.9%, and specificity of 87.1%.

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	Graph convolutional network, penyakit, PPOK, sensor gas, COPD, disease, gas sensor
Subjects:	Q Science > QA Mathematics > QA336 Artificial Intelligence R Medicine > R Medicine (General) > R856.2 Medical instruments and apparatus. R Medicine > R Medicine (General) > R858 Deep Learning T Technology > TA Engineering (General). Civil engineering (General) > TA1573 Detectors. Sensors
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55101-(S2) Master Thesis
Depositing User:	Dava Aulia
Date Deposited:	01 Feb 2024 04:50
Last Modified:	01 Feb 2024 04:50
URI:	http://repository.its.ac.id/id/eprint/105876

Actions (login required)

View Item