Early detection of Myocardial Infraction by Blind Source Separation of ECG Signals

The objective of this project is to try and manipulate the ECG signals that we have in order to achieve an easy way for identifying heart disease.

Abstract
The objective of this project is to try and manipulate the ECG signals that we have in order to achieve an easy way for identifying heart disease. A prominent and fatal heart disease is called Myocardial Infraction, more commonly known as “Heart Attack”. Modern medicine can identify the heart attacks in the majority of cases, most of them only after they have occurred. Our project is a step in the goal for identifying heart attacks in an automated way, and the result may provide a tool for identifying heart attacks before they happen. This project will try to give an answer for: “Is there a way to identify heart attacks by looking at the separated sources of the ECG signal?”

 

What is ECG?
ECG (or EKG) stands for electrocardiogram; which is a measurement of the electrical signals that control the rhythm of the heartbeat. The heart is a muscular organ that beats in rhythm to pump the blood through the body. The signals that make the heart’s muscle fibers contract come from the senatorial node, which is the natural pacemaker of the heart.

The shape of the ECG: 1

 

 

What is a Myocardial Infarction?
“Myocardial Infarction” (abbreviated as “MI”) means there is death of some of the muscle cells of the heart as a result of a lack of supply of oxygen and other nutrients. This lack of supply is caused by closure of the artery (“coronary artery”) that supplies that particular part of the heart muscle with blood. This occurs 98% of the time from the process of arteriosclerosis (“hardening of the arteries”) in coronary vessels.

 

Blocks diagram

2

 

Results – QRS signal
We have a signal that appears in both ECG; normal and sick ECG. This signal can be thought on as representing the RQS signal in the ECG. The most obvious thing we see immediately is that the signal itself is very “clean”, it does not have ripples.
We found out that the similarity between the sick ECG and the normal ECG in the QRS signal gets better with the time, in other words the similarity of the last sample of the sick ECG is better than the similarity of the first sample of the sick ECG. The results:

3 4

 

Results – P and T signals
We have another signal that appears in both ECG; normal and sick ECG. This signal can be considered as the signal that represents the P and T signals in the ECG. The difference between this signal and the QRS signal is that we don’t see improvement in the sick ECG in three hours as we saw in the QRS signal, therefore we can conclude that this signal will take more time to return to it’s “normal” shape. This signal can be used to detect smaller heart attacks.

5 6

 

Results – The signal number
The last comparison result is the number of signals in the sick result that appears to be similar to the normal result signals. In the beginning we have 3 common signals and in the end we have 4 common signals:

7 8

The important point in this result that, we can’t find such result in the original ECG, cause in the original ECG we have 12 signals and that’s it. But here we could detect sick ECG using the number of signals that is similar to the normal ECG.

 

Future work
Our project is only one step in the path for predicting heart attacks. As have been seen from the results before we can identify heart attacks using blind source separation method, and in one of the results we got a way for identifying small heart attacks. But much work still lies in front of us to reach the ultimate goal – predicting heart attacks.

  • A bigger and longer research has to be made in order to validate the results that we found. This require a bigger database than what we have and can be integrated in hospitals in addition to the traditional methods, and with time it can be proved as a valid method or not, and also it can be improved this way
  • Another research can be made to try and test the method on another kind of heart diseases, which might have predicate symptoms which are difficult to be identified in the traditional ECG
  • Building an automated heart attack identifier, which would identify heart attack (or other diseases) according to the parameters of the signal we got