Neural network classification of body surface potential contour map to detect myocardial infarction location

This paper is the follow-up of our previous work presented for CinC/PhysioNet Challenge 2007 on the “electrocardiographic imaging of myocardial infarction”. We have presented an automatic method for MI location detection by Neural Network classification of BSPM data. Data used here contain BSPM signal of four patients and their actual infarcted segments (two training cases and two cases for test). By mapping Q-wave integral and QRS-complex integral on torso surface and applying four threshold-based rules, an abnormal area on the torso can be obtain. This detected abnormal area then is mapped to the heart segments. ANN classifier is used at final step. The results expressed by parameter OS (overlapped segment) which is a value between 0 and 1, where 1 is a perfect match. The results for two test cases are OS_case#3=0.7 and OS_case#4=0.4 shows this mathematically simple method can predict the location of MI reasonably. However further works is needed to improve the results.