Resnet u-net decoder improves ecg wave boundary detection accuracy
Decoder Design Matters for ECG Delineation
Machine LearningArtificial Intelligence
Summary
Identifying the start and end of important heart signals in an ECG helps computers better interpret heart health. Usually, training such models takes a lot of expert-labeled data, which is hard to get. The authors show that using a specific combination of neural network parts—called ResNet-18 as the encoder and U-Net as the decoder—improves accuracy more than previous methods. Their work also highlights that how the decoding part of the model is designed matters more than the semi-supervised learning techniques used before.
What this means in practice
- •For medical device developers: Improve automated ECG analysis tools by integrating more accurate ECG wave boundary detection from the proposed ResNet-U-Net model.
- •For clinical data engineers: Build workflows for annotating ECG datasets with higher precision using improved model architectures for wave delineation.
Authors
Joseph Scharpf, William Han, Chaojing Duan, Michael A. Rosenberg, Emerson Liu, Ding Zhao
Abstract
Electrocardiogram (ECG) delineation identifies the boundaries of P waves, QRS complexes, and T waves, providing structural annotations that can guide AI models in learning to interpret ECGs. However, training accurate delineation models requires manual annotations that are scarce and time-consuming to obtain. Recent work addresses this limitation through semi-supervised learning (SSL), but the design of the architecture, particularly the decoder, has received less attention. To this end, we propose R-U-Net, an ECG delineation model that pairs a ResNet-18 encoder with a U-Net decoder. On SemiSegECG, R-U-Net outperforms the strongest evaluated ResNet-18 + fully convolutional network (FCN) head baseline in each of the 16 in-domain settings by 3.3-13.0 mIoU and achieves 82.6 mIoU in the cross-domain setting, an improvement of 8.1 mIoU. Controlled ablations show that decoder design contributes more to performance gains than the evaluated SSL methods, motivating further exploration of architectures for ECG delineation. All code is open-source at github.com/ELM-Research/ECG-Delineation.