Abstract:
The main goal of this work was to design a method for automatic detection of
various respiratory diseases using audio recordings. Prior attempts at automated
classification of adventitious respiratory sounds have tried to simplify the problem by
focusing on a single type of sound and, to the best of our knowledge; single sound
detecting systems had very good accuracy (up to 96%) however they could only be
used for detection of one particular disease. Our goal was to build a system that can
accurately classify 7 different types of diseases and able to tell if a person is healthy.
The last recorded system built for this purpose was in March 2018, which had 67.077%
accuracy and was able to detect both types of sounds, wheezes and crackles.
The expected end result for this project is to use alternative methods, never tried
before, to increase the accuracy ofthe previously built systems. For this purpose, we
have acquired the same dataset used in previous versions of this model and created a
model that has significantly better accuracy.
The motivation for this project was to help the doctors and test centers for
diagnostics of such diseases and help them in their decision making. Many people
lose their lives because ofmisdiagnosed diseases due to doctor’s lack of experience,
equipment malfunction and other external factors (i.e noise while using the
stethoscopes). This system will provide support to the doctors in diagnosing these
diseases accurately.
As a result, we can save precious lives and diagnose respiratory diseases
effectively
