Firstly, the database about MI detection is PTB database.
I suggest that you can read the introduction part of the paper entiled
automated interpretable detection of myocardial infarction fusing energy entropy and morphological features, which lists lots of papers about MI detection.
Secondly, you may mistake the classification of MI according to your last reply. The diagnostic result
for MI or healthy control is based upon the 12 leads ECG records, not the beats. So the annotation beat by beat is not necessary.
Finnaly, deep learing methods based on CNN LSTM and combined approaches are more and more popular, you can try it based on inter-patient mode.
Teimoor Bahrami I know you want to employ the supervised learning algorithm. For example, the input is 12*8000 (12 is the number of lead, 8000 indicated the sample points every lead), the output is the label 1 or 0 (1 for MI and 0 for HC). You can extract the features based on wavelet transform, entropy or other methods, meanwhile, train the model to detect MI based on SVM, RF or CNN....