Automated T-wave Delineator Algorithm for J-t^sub Peak^ And T^sub Peak^–T^sub End

Introduction: Two recent clinical trials demonstrated that combined analysis of the QT sub-intervals J-Tpeak and Tpeak–Tend can differentiate drugs that selectively block hERG potassium channels (high torsade de pointes risk) from drugs that block hERG and late sodium or calcium currents (low torsade de pointes risk). In this abstract, we describe an automated T-wave delineator algorithm for marking Tpeak and Tend in the vector magnitude lead to enable fully-automated measurement of J-Tpeak and Tpeak–Tend. Methods: The automated T-wave delineator algorithm consists of three phases: 1) finding the point of interests called candidates, 2) applying pre/ post processing rules on candidates to find Tpeak and Tend, and 3) readjusting Tend by minimizing a cost function in a vector magnitude lead. In the first phase, candidates were selected from the local maxima of the first derivative of the vector magnitude lead. Each candidate has three main properties of peak, maximum rising slope, and minimum falling slopes for finding Tend. In the second phase, sequential rules, extracted from the vector magnitude in a clinical trial were applied to trim the candidates by removing a candidate or merging two candidates together. In the last phase, Tend was readjusted by optimizing Ƒ(e,d) in which d is the ratio between Tend and heart rate and e is the ratio between Tend amplitude and Tpeak amplitude. The performance of the automated T-wave delineator was compared with delineations from two human annotators in two clinical trials. For this comparison, delta differences between raw data and single delta from baseline (Δb) were determined to assess bias. Results: The delta differences between raw data of automated algorithm for the first and second clinical trials, respectively, were ΔTend,1 ≈ −11.4 ± 2.90 ms, ΔTpeak,1 ≈ 1.20 ± 2.70 ms, and ΔTend,2 ≈ −5.22 ± 2.97 ms, ΔTpeak,2 ≈ −0.26 ± 9.26 ms. The higher variation of ΔTpeak,2 is due to disagreement between the presence of notches in 57 T-waves (1.4% of total) in first clinical trial. Single delta measurements from baseline for heart rate-corrected J-Tpeak (J-Tpeakc) and Tpeak–Tend between the automated algorithm and the clinical trials data were Δ(Δb Tpeak-Tend,1) ≈ −0.97± 2.97 ms, Δ(Δb J-Tpeakc,1) ≈ 0.16 ± 2.27 ms and Δ(Δb Tpeak–Tend,2) ≈ − 0.00 ± 8.65 ms, Δ(Δb J-Tpeakc,2) ≈ 0.42 ± 8.70 ms. Discussion: The results show consistency between automated T-wave delineator and delineations made by two human annotators. The main disa