Sensi uses a combination of time and frequency features from the heart sound to determine if the recording is pathological or innocent/normal, in the same way a trained physician will use his/her knowledge and skill.
Sensi was developed using a large (echo) validated set of heart sounds, pathological and normal. Time and frequency domain features are extracted from each heart beat, fed into a artificial neural network that was trained with the above mentioned data set of nearly 2000 heart sounds to perform the classification.
During analysis the following steps are followed:
- For each heart sound, all the S1 and S2 sound pulses are identified.
- A set of time domain features are extracted from each heart beat.
- Each heart beat is divided into a set of "bins".
- A short term Fourier transform is calculated for each "bin".
- Frequency domain features are extracted from each "bin".
- This set of features are fed into a neural network that will classify the heart beat as either pathological or normal.
The next image shows a representation of the time domain features and the bins into which each heart beat is divided.
The next image shows the corresponding frequency domain representation of the heart beat's systolic area.
It should be noted that a neural network classifier was developed for each heart sound recording location. This is due to the sound variations for each recording location.