Multiclass Classification of Respiratory Diseases From Stethoscopic Lung Sound Signals Using Deep Learning

Hidayah, Rohmah (2024) Multiclass Classification of Respiratory Diseases From Stethoscopic Lung Sound Signals Using Deep Learning. Masters thesis, Institut Teknologi Sepuluh Nopember.

Text 6022231110-Master_Thesis.pdf - Accepted Version Restricted to Repository staff only until 1 April 2027. Download (8MB) | Request a copy

Abstract

Respiratory diseases are one of the leading causes of death worldwide, including Asthma, Bronchitis, COPD (Chronic Obstructive Pulmonary Disease), Heart Failure, LRTI (Lower Respiratory Tract Infection), Lung Fibrosis, Pleural Effusion, Pneumonia and URTI (Upper Respiratory Tract Infection). Auscultation is often an alternative for doctors to diagnose these respiratory diseases. However, this method is highly dependent on the knowledge and experience of each doctor. Therefore, a consistent and standardized auscultation method is needed to improve the accuracy of respiratory disease diagnosis. Previous research using the Adaptive Neuro-Fuzzy Inference System (ANFIS) algorithm achieved 94.94% accuracy, 89.87% specificity, and 100% sensitivity, only for the classification of two classes, namely asthma and normal. Unlike previous studies that used one dataset, in this study two datasets were used. The primary dataset was obtained as part of an ongoing project at King Abdullah University Hospital, Jordan University of Science and Technology, Irbid, Jordan. The second dataset was taken from the publicly available ICBHI Challenge database to complement the main dataset. After success in previous studies, in this study, we propose the use of varied Deep Learning methods to perform multiclass classification of ten types of respiratory diseases. The objective of this study is to classify 10 respiratory diseases by combining time domain and frequency domain to find out the effective method for classification. The Deep Learning proposed in this study are CNN, LSTM, CNN-LSTM, ANFIS, ANFIS-LSTM. Experimental findings show that combining the frequency domain and time domain in deep learning methods results in higher accuracy, sensitivity, specificity and precision compared to previous studies.
=================================================================================================================================
Penyakit pernapasan merupakan salah satu penyebab utama kematian di seluruh dunia, termasuk Asma, Bronkitis, COPD (Chronic Obstructive Pulmonary Disease), Gagal Jantung, LRTI (Lower Respiratory Tract Infection), Fibrosis Paru paru, Efusi Pleura, Pneumonia dan URTI (Upper Respiratory Tract Infection). Auskultasi sering menjadi alternatif para dokter untuk melakukan diagnosis terhadap penyakit pernapasan tersebut. Meskipun begitu, metode ini sangat bergantung pada pengetahuan dan pengalaman masing masing dokter. Oleh karena itu, metode auskultasi yang konsisten dan terstandarisasi diperlukan untuk meningkatkan akurasi diagnosis penyakit pernapasan. Penelitian sebelumnya menggunakan algoritma Adaptive Neuro-Fuzzy Inference System (ANFIS) mencapai akurasi 94.94%, spesifisitas 89.87%, dan sensitivitas 100%, hanya untuk klasifikasi dua kelas yaitu asma dan normal. Tidak seperti Penelitian sebelumnya yang menggunakan satu dataset, dalam penelitian ini menggunakan dua dataset, Dataset utama diperoleh sebagai bagian dari proyek yang sedang berlangsung di King Abdullah University Hospital, Jordan University of Science and Technology, Irbid, Yordania. Dataset kedua diambil dari database ICBHI Challenge yang tersedia secara publik untuk melengkapi dataset utama.Setelah berhasil dalam penelitian sebelumnya,dalam penelitian ini, kami mengusulkan penggunaan metode Deep Learning yang bervariasi untuk melakukan klasifikasi multikelas sepuluh jenis penyakit pernapasan. Tujuan dari penelitian ini adalah untuk mengklasifikasikan 10 penyakit pernapasan dengan menggabungkan domain waktu dan domain frekuensi guna mengetahui metode yang efektif untuk melakukan klasifikasi. Deep Learning yang diusulkan dalam penelitian ini CNN, LSTM, CNN-LSTM, ANFIS, ANFIS-LSTM. Temuan eksperimental menunjukkan bahwa dengan metode menggabungkan domain frekuensi dan domain waktu pada metode deep learning menghasilkan akurasi, sensitivitas,spesifisitas dan presisi yang lebih tinggi dibandingkan dengan penelitian sebelumnya.

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	Respiratory Disease Classification, Deep Learning, Multiclass classification, Combined Frequency Domain and Time Domain, CNN, LSTM, CNN-LSTM, ANFIS, ANFIS-LSTM
Subjects:	T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK5102.9 Signal processing.
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Electrical Engineering > 20101-(S2) Master Thesis
Depositing User:	Rohmah Hidayah
Date Deposited:	07 Jan 2025 02:32
Last Modified:	07 Jan 2025 02:32
URI:	http://repository.its.ac.id/id/eprint/116182

Actions (login required)

View Item