Title :
QRS detection and cardiac cycle separation without ECG
Author :
Aase, Svein Arne ; Snare, Sten Roar ; Mjølstad, Ole Christian ; Dalen, Håvard ; Orderud, Fredrik ; Torp, Hans
Author_Institution :
Dept. of Circulation & Med. Imaging, Norwegian Univ. of Sci. & Technol. (NTNU), Trondheim, Norway
Abstract :
Automatic detection of the QRS complex on ECG is used on most cardiac ultrasound scanners to separate recorded ultrasound data into cardiac cycles for continuous playback and storage. On small hand-held scanners it is unpractical to connect ECG cables. We aim to do automatic cardiac cycle separation and detect the start-point of a cardiac cycle corresponding to QRS on ECG using only B-mode ultrasound data. A deformable non uniform rational B-spline (NURB) curve is fitted to apical 2D B-mode data using an extended Kalman filter framework. The model is updated using a combination of edge detection and block matching. Resulting displacement of the mitral annulus of the left ventricle is extracted and processed by a template-based curve analysis algorithm determining whether the curve has the shape expected of mitral annulus displacement curves. If it does, the first time point of maximum distance from the probe is detected. A section of the displacement curve before and after this maximum is extracted and compared to the rest of the curve by a sum of absolute differences (SAD) algorithm to find the point where the curve repeats itself, the cycle length. The parameters and rules of the algorithm were trained using a data set of 105 apical B-mode images with 3 cardiac cycles and ECG. Testing the algorithm against ECG cycle start and cycle length was done on a separate data set of 82 apical images with 3 cardiac cycles. To emulate a real-time situation, several test cases with different start point during the first cardiac cycle were run for each recording. Totally, 4655 test cases were run. In 935 cases (20%), the curve shapes were automatically discarded. In remaining cases (80%), mean ? SD difference of cycle start against corresponding ECG QRS trig point was -24 ? 64 ms. Cycle length was detected in 3664 cases (79%of total). Mean ? SD difference against corresponding cycle lengths by ECG was 3 ? 35 ms. Automatic detection of cardiac cycle lengths and time points near - QRS of ECG is feasible based on B-mode only.
Keywords :
echocardiography; edge detection; electrocardiography; medical image processing; medical signal detection; splines (mathematics); ECG; QRS detection; automatic cardiac cycle separation; block matching; deformable nonuniform rational B-spline curve; displacement curve; edge detection; left ventricle mitral annulus; sum of absolute differences algorithm; template-based curve analysis algorithm; ultrasound scanners; Cables; Data mining; Electrocardiography; Shape; Spline; Storage automation; Surface reconstruction; Surface topography; Testing; Ultrasonic imaging;
Conference_Titel :
Ultrasonics Symposium (IUS), 2009 IEEE International
Conference_Location :
Rome
Print_ISBN :
978-1-4244-4389-5
Electronic_ISBN :
1948-5719
DOI :
10.1109/ULTSYM.2009.5441469