Skip to content

Latest commit

 

History

History
40 lines (24 loc) · 2.52 KB

File metadata and controls

40 lines (24 loc) · 2.52 KB

Determine heart rate from a digital stethoscope using Machine learning

I worked Sensordrop Networks (http://www.sensordropsnetworks.com/) from 20th July,2020 to 20th Sept,2020 as an intern. During this time I worked with Heart beat signals as .wav format. My work was to build a statistical model to predict heart rate from this signals. Also, I tried to develop an accurate peak detection algorithm.

In this project we worked two different heart beats dataset

Time domain visualization of the Heart beat audio signal before and after Bandpass filtering

image image

MFCC features plot of the Heart beat audio signal before and after Bandpass filtering

image image

Actual vs Predicted values of Heart beats in my test set

image

Peak detection part

Actual signal before and after Band pass filtering and avereging

image image

Detected peaks but here we can see some unwanted peaks

Try to remove unwanted peaks. Here I used two metods.

  • thresholding: set the threshold value maximum of signal value multiply with a factor 0.1

  • distance: I consider an original peak should be have maintain a distance 1500 from it's previous and next peak

image

After filtering peaks

image

Plot the time interval between two peaks

image