Performance evaluation for the sliding area-based T wave detection method on the QT database

The accurate detection of T-wave characteristic points is the basis of the morphology judging, which plays a great role in the diagnosis of heart diseases, especially in the arrhythmia diagnosis. Recently, the method based on the sliding area has received increasing interest due to its stable robustness. In this study, we aim to evaluate the performance of T-wave detection based on the sliding area method. Firstly, we constructed a systematic evaluation scheme by using records from the QT database and records chosen were separated into 10 groups, randomly. Then, the sliding area method was evaluated on each group. The sensitivity, specificity, accuracy and F1 measure were used as the evaluation indexes. Additionally, the tolerance window for determining a correct segmentation was 100ms. In this study, 1371 beats were used to detect the T-wave onsets. And for 1371 beats, the values of Se, P+, Acc and F1 measure were 54.70%, 54.70%, 37.66%, 54.70% for signals of first channel and were 54.05%, 54.05%, 37.03%, 54.05% for signals of second channel. Besides, 3542 beats were used to detect the T-wave offsets. And for 3542 beats, the values of Se, P+, Acc and F1 measure were 87.83%, 87.83%, 78.30%, 87.83% for signals of first channel and were 86.73%, 86.73%, 76.57% 86.73% for signals of second channel. Values of index aforementioned were smaller than 90%, which illustrated that the performance of the sliding area method depends on its different parameters combination and it adaptability needs to be verified further.