An Efficient Low Computational Cost Method of R-Peak Detection

Visual inspection of R-peaks in an electrocardiogram (ECG) signal may lead to wrong diagnosis due to physiological variability and the noisy status of the QRS complexes causing its incorrect interpretation. Hence, computer-aided diagnosis (CAD) is required for better and correct diagnosis of cardiovascular diseases and interpretation of essential clinical information. Low computational cost CAD systems with good accuracy are always preferred in health informatics. Presently, performance of most of the ECG signal analysis techniques such as feature extraction, classification etc. depends heavily on the used pre-processing technique, which may consume appreciable portion of the total processing time. Therefore, it can be inferred that if pre-processing technique is skipped altogether without compromising the detection accuracy of a diagnosis, then it would save central processing unit (CPU) time reducing the overall computational time. This time saving will turn out to be very important in critical and emergency situations to save lives of several patients. Hence, the authors were motivated to propose an efficient low computational cost method based on fractional Fourier transform (FrFT) as a feature extraction technique eliminating the need of any pre-processing. In other words, the proposed technique is applied directly on the raw ECG data resulting in computational savings. Firstly, eigenvalues and eigenfunctions are proposed to be obtained using FrFT. And secondly, these are proposed to be utilized for R-peak detection using Independent Principal Component Analysis (IPCA) on the basis of kurtosis and variance of the extracted features in a noisy ECG signal with different morphologies. The proposed methodology is evaluated on the basis of various performance parameters viz. sensitivity (SE), accuracy (ACC), and positive predictive value (PPV) (of the detected ECG beats). SE of 99.97%, PPV of 99.98%, and ACC of 99.94% are obtained on MIT-BIH Arrhythmia (MIT-BIH Arr) database. All simulations are done using Intel Core i3-3240 Dual-Core Processor 3.4 GHz and 8 GB of RAM using MATLAB R2011a. The average processing time of CPU is observed to be 0.677 s with detection error rate (DER) of 0.058%. Both these values are least among other existing techniques, which establish that the proposed method incurs low computational cost. Also, consistently high values of all the performance parameters such as SE, PPV and ACC demonstrates the robustness of the proposed technique. Hence, the proposed methodology is expected to assist cardiologists for intelligent, effective, and timely diagnosis of heart rhythm irregularities.