Abstract:
Medical field is becoming more and more challenging due to large number of heart problems faced by the people. Almost all doctors are always trying their best to give a best treatment to a patient so that the patients get recovered from the heart problem. The heart patient is carefully diagnosed by analyzing ECG signals. Many researchers have proposed different distance measures to compare ECG trajectories for similarity. Dynamic Time Warping(DTW), Edit Distance with Real Penalty (ERP), Edit Distance with Real Sequence (EDR), Longest Common Subsequence (LCSS) are used to compare time series trajectories for similarity. The existing distance measures are computing similarity distance based on a proximity distance. There are three main drawbacks faced by the existing distance measures. The first drawback is, ECG trajectories are not compared based on a shape feature. The second problem is, existing edit distance measures are not rotation invariant. The third problem is, they are very sensitive towards noise. Due to these problems, the performance of the existing distance measures is comparatively low. In this paper, we have proposed Turning Function Based (TFB) distance measure to compares ECG trajectories based on a shape feature. TFB distance measure is based on a angular distance of the ECG trajectories. The angular distance captures the shape feature of the ECG trajectory. We have carried out experimental study on the real time and synthetic datasets. Experimental results reveal that our proposed TFB distance measure compares the trajectories based on the shape feature. Further, our experimental results reveal that, TFB distance measure supports rotation invariant property and very robust to the noise.
