FFCResNet Achieves First Large-Scale ECG Interval Accuracy Study

A new study on arXiv introduces an advanced system that automatically measures ECG intervals and identifies wave patterns, using a technique called Fast Fourier Convolution ResNet (FFCResNet). It analyzes 10,646 clinical 12-lead ECGs, marking the first detailed look at measurement accuracy, complete with statistical evaluations like bias and 95% limits of agreement. The FFCResNet combines local temporal convolution with global spectral analysis via FFT and uses register tokens to improve feature learning. The research develops three models for delineating P, QRS, and T waves. This work is significant as it addresses a lack of large, diverse datasets in current literature, enhancing the accuracy of cardiac diagnostics for intervals like PR, QRS duration, and QT/QTc.