Register for free to
view the full innovation profile.
Novel approach for automatic ECG delineation and QT interval estimation, using an unsupervised learning model to diagnose and classify LQT syndrome.
About
BACKGROUND ECG segmentation and QT interval estimation are essential in ECG-based diagnosis of long QT syndrome (LQTS). LQTS is a heart rhythm disorder, associated with abnormal ventricular depolarization/repolarization or contraction of different heart chambers. LQTS can result in an abnormally fast heart rhythm called torse de pointes which is associated with sudden cardiac death. It is clinically characterized by a prolonged QT interval and abnormal T wave morphology. The prevalence of congenital long-QT syndrome is estimated at 1 in 2000 live births. Accurate measurement of peak, onset and offset P wave, QRS complex and T wave is required in advanced automatic ECG interpretation systems. Current ECG delineation models suffer from a variety of issues such as being sensitive to low-amplitude T and P waves, noise and baseline wandering. There is a need for a comprehensive automatic approach to deal with different morphologies and amplitude levels of T and P waves in noisy ECGs with baseline wandering. TECHNLOLGY OVERVIEW Queen’s researchers have proposed a novel approach which delineates different waves of ECG signals using an unsupervised clustering algorithm to improve the accuracy of QT interval measuring. The method identifies a set of features in the data based on inflection points (IP) of the ECG signal. The IPs are located using a LoG filter which makes the algorithm robust to noise and baseline wandering. ECG signals are segmented to the waves between two consecutive IPs. A 3D feature space is created based on truncated energy, width, and sum of slopes at their boundaries and an EM algorithm for Gaussian mixtures is applied. Experimental results showed that the algorithm extracts ECG waves accurately, even if they have low energy and amplitude.
Key Benefits
• Validation studies demonstrated on overall Long QT test accuracy of 91.3% with sensitivity = 91% and specificity = 92%. • The new method enables detection of Long QT but also effectively classifies Long QT type 1 and 2 (represents > 65% of Long QT patients) which currently need genetic testing. Currently up to 25% of LQTS cases remain genotype elusive. Validation of the classification detection algorithms had an overall test accuracy of 94% with sensitivity = 89% and specificity = 100%. • The algorithm is robust to noise and baseline wandering. • There is no preprocessing window approach guaranteeing a higher resolution and lower computational complexity.
Applications
Diagnosis of Long QT Syndrome and classifying types of Long QT 1 and 2