Conclusion
We were able to demonstrate the classification of different activities with the ECG & SCG signal with an accuracy of ~93%. We also demonstrated a combined feature accuracy of 85%.
According to feedback from our survey, our device has an easy to use app. The device is very portable and hardly noticeable except in specific movements. The subjects were also unconcerned with privacy and thought the test was more thorough than a doctor's physical.
Further Avenues of Exploration
-
We were limited on our subject population size. We would like to repeat this study with more subjects.
-
We would also like to expand the length of the experiment to see if our device could actually become a long-term solution. We would like to target a timespan of a few days.
-
We could also use other accelerometer datasets to train and improve our accelerometer classification.
-
Perform better coding algorithm and pipelining.
-
Leverage more data that is collected instead of just the accelerometer, ECG and SCG signals like heart rate, respiratory rate and blood pressure.
-
Possibly use deep learning to use the continuously collected data to infer a much more complex and difficult diagnosis like stress or tracking heart adaptations from daily life.