QRS complex detection based on simple robust 2-D pictorial-geometrical feature

Abstract

In this paper a heuristic method aimed for detecting of QRS complexes without any pre-process was developed. All the methods developed in previous studies were used pre-process, the most novelty of this study was suggesting a simple method which did not need any pre-process. Toward this objective, a new simple 2-D geometrical feature space was extracted from the original electrocardiogram (ECG) signal. In this method, a sliding window was moved sample-by-sample on the pre-processed ECG signal. During each forward slide of the analysis window an artificial image was generated from the excerpted segment allocated in the window. Then, a geometrical feature extraction technique based on curve-length and angle of highest point was applied to each image for establishment of an appropriate feature space. Afterwards the K-Nearest Neighbors (KNN), Artificial Neural Network (ANN) and Adaptive Network Fuzzy Inference Systems (ANFIS) were designed and implemented to the ECG signal. The proposed methods were applied to DAY general hospital high resolution holter data. For detection of QRS complex the average values of sensitivity Se = 99.93% and positive predictivity P+ = 99.92% were obtained.

• Adaptive network fuzzy inference system
• classification
• feature extraction
• K-nearest neighbors classification
• QRS complex detection-delineation