Our Results

     for the parsing filter indicate that parsing is best suited for semi-regular heartbeats whose periodicity does not change quickly, frequented by itermittent but sometimes high-power noise. With correctly identified parameters, the parsing filter could do hours of low-noise ECG without an error, but would likely fail when the signal-to-noise-ratio (SNR) would remain close to 1 for more than a few heartbeats.

Unfortunately the parsing filter was somewhat sensitive to parameters, which meant that automatic analysis was initially out of the question. Later we implemented a linear boxcar-like filter to detect and discard low-SNR ECG, as well as making the analysis automatic. However, this automatic filter was not more successful, but this lack of success had more to do with the short amount of time that we had to implement and test it. In general we were pleased with what we achieved with algorithmic parsing but we were also frustrated with how the simple problems of the algorithm that remained seemed almost impossible to circumvent.