Detect Ventricular Fibrilation

This database is the result of a project that aims to present a new algorithm for detecting ventricular fibrillation , exploration and evaluation of the combination of different ECG metrics and their discrimination of the disease.

The choice for the ECG's fell on the CU Ventricular Tachyarrhythmia Database available in website Physionet .This database includes 35 ECGs from patients who have experienced periods of ventricular tachycardia and ventricular fibrillation. The ECG of each individual was acquired with a sampling frequency of 250 Hz for about 8.5 minutes corresponding to 127 thousand and 232 samples.

Thus, from the original database, we chose 25 subjects in which each one of them has of at least 10,000 samples (40s) for each episode of fibrillation and without fibrillation (group control). After , the entire algorithm was designed comprising in its process: the filtering of the signal, the respective selection and extraction of metrics, the choice of time windows for calculating the metrics, their evaluation and still the methodology for classification.

PRE-PROCESSING:

However, for a correct ECG analysis it is necessary to perform a pre-processing of it before going on to a more in-depth study. So we removed power line interference and we correct the baseline of the ECG.

FEATURE EXTRACION:

FREQ.F:
The fundamental frequency is the frequency that is predominant in a certain signal (spectral domain), that is, it is that which is stronger and which contributes the most to the power density of all the harmonics that make up, in our case the ECG.

POWER (Potencias):
The last metric we extract from the spectral analysis of our signal is power.The power of the ECG can be calculated again from the spectral density of the signal. As the spectral density reveals the power that each frequency contributes to the signal, if we add the contribution of all frequencies[15], we will have the total power of the analyzed segment. So the total power is no more than the area of ​​the entire power spectral density.

N(%) - Percentage of samples that exceed a V0 value:
The two new metrics presented are obtained through an analysis of the ECG in the temporal domain. Now, the metric presented here is the parameter N of the TCSC algorithm .

F1-Ratio:
The last metric used is perhaps the most atypical and least common compared to metrics described above. This metric is based on the extraction of characteristics from two graphs made using the ECG.
In the first step, a first graph is developed in which X axis are the samples of the analyzed ECG (s1, s2, s3,…,si.) and the ordinate (y) are the same samples with advances of a sample(s2,s3,s4,s(i+1)). The elaboration of the second graph is quite similar to the first, however, for the latter we placed an advance of 5 samples in the ordinate (s(i+5)) compared to the x axis.
We convert this graphs to a binary image and the F1 metric is obtained by subtracting the black pixel ratio for the graph with 5 early samples (ratio 2) by the ratio of black pixels for the graph showing 1 lead sample (ratio 1).

CONCLUSION:

The dataset below is the result of this project for 25 patients that experienced the 2 heart condiction (so 50 rows) and i did all this job of processing sinal in MATLAB . I hope that you can take profit of this dataset for future works and evaluate what is the best model of machine learning that lead us to better detect the pathology This can be usefull for apply in a cardiac disfibrilator and prevent deaths.