Treffer: Rhythmic analysis of human heart sounds applying deep learning: LSTM and CNN.

The heart is a vital part of our existence and to identify issues pertaining to any heart related fatalities are crucial in our modern world. Deep Learning and Machine Learning in the domain of Artificial Intelligence helps us in the early detection and accurate diagnosis based on the learnings obtained from previous real time datasets. Using the audio datasets of heart sounds from clinical trials as well as public data, we have aimed to create models utilizing deep learning methods like long short term memory network and convolutional neural network. We have used Mel frequency cepstral coefficients to extract features from the audio files making use of inbuilt functions from the librosa library in python. Model validation is used to evaluate these algorithms and to obtain higher accuracy. An accuracy of 96.5811% and 92.3076% had been achieved using LSTM and CNN respectively by implementing a band filter, the additional noise apart from the heartbeat sound signal which is the 'lub' and the 'dub' is removed and a size is fixed for the sampling rate of each sound signal. The implementation process and analysis has been conducted using the python programming language in Google Colab notebook to make use of the several imported libraries. [ABSTRACT FROM AUTHOR]