Detection of Chronic Lung Disorders using Deep Learning

Abstract

Lung disorders can be fatal if not treated in the right manner. Symptoms of respiratory disorders include wheezing, breathlessness or difficulty in breathing, cough, hoarseness, and chest pain to name a few. Although the symptoms look common, lung disorders often go undetected due to various reasons such as misdiagnosis, expensive diagnostic techniques, lack of awareness and negligence. In most cases, the patient is required to take a pulmonary function test which includes thoracoscopy, chest imaging (X-rays), electrocardiography and bronchoscopy. In this paper we explore an alternative technique for detecting respiratory disorders through analysis of lung sounds. Lung sounds are vital factors of respiratory health and disorders. They are produced due to the movement of air and secretions in lung tissue or they might also be generated due to the presence of any infection or anomalies. Asthma or Chronic Obstructive Pulmonary Disease (COPD) patients often wheeze as a result of an obstructive airway disease. These sounds can be captured using digital stethoscopes which can then be converted into audio signals for further processing. This audio data gives us the opportunity to diagnose respiratory disorders like pneumonia, asthma and bronchiolitis using deep learning techniques such as convolutional neural networks. We also propose a design for the digital stethoscope which can help record lung audio samples.