کنفرانس مهندسی برق ایران

صفحه اصلی / سی و یکمین کنفرانس بین المللی مهندسی برق

BLSTM-Convolutional Neural Networks for Respiratory Disease Diagnosis

نویسندگان :

Mohammad Hassan Khamechian¹ Mohammad Reza Akbarzadeh Tootoonchi²

1- دانشگاه فردوسی مشهد 2- دانشگاه فردوسی مشهد

کلمات کلیدی :

respiratory diseases،convolutional neural network،BLSTM-CNN،Audio features

چکیده :

Even before the coronavirus, respiratory illnesses could not be neglected. These diseases are responsible for a sizeable fraction of annual global population deaths. Numerous and diverse respiratory illnesses exist. Subtypes of this illness include chronic obstructive pulmonary diseases, respiratory cancers such as lung and laryngeal malignancies, respiratory tract infections, and coronavirus. This project suggests combining convolutional neural networks with bidirectional long short-term memory. The suggested approach is superior to other contemporary papers since it accurately (average of 92% accuracy) identifies more respiratory disorders (6 respiratory diseases and healthy peaple). In recent years, because of the high precision and noise resistance of convolutional neural networks, they have been utilized in a variety of applications, including signal and image processing. Furthermore, The BLSTM approach is the most intelligent way to solve time series challenges since it saves the dependencies of input sequences in models and can deal with difficulties including vanishing gradients. Therefore, a combination of these two approaches has been used to detect some respiratory disorders using the audio respiration signal from a digital stethoscope. This article also uses data augmentation and filtering to create more data as the preprocessing methods. The ICBHI'17 database, the richest and most comprehensive collection of respiration sound signals available to the public, serves as the foundation for this investigation.