Tuning a real-time detector of transient cardiac ischemic episodes on the Long-Term ST database according to the annotation protocol B

This study revisits a real-time ischemia detector developed over the Long-Term ST (LTST) database according to the transient ST episode annotations for protocol A. We have studied how to adapt the initial detector in order to comply with the ischemia annotation protocol B requirements on the same database, a protocol similar to what is being used in clinical practice. Secondly we have studied to improve the initial algorithm. The intermediate results in terms of sensitivity/positive predictivity after a first detection step were 83.15%/56.75% and improved to 81.17%/74.83% after a discrimination step based on the bagging of decision tree method. These results improve our previous ones and are comparable to other results reported for this database, with the advantage that this algorithm could analyze the ECG signal in real-time.