ScholarWorks@동국대학교

초록

Manual interpretation of ECG data faces complex challenges because of difficult ECG waveforms that undergo morphological variations. The accurate identification and segmentation of signals becomes more difficult because of noisy signals and unbalanced datasets. The proposed SkipCNN based ECG Transformer Spike model represents a new deep learning framework that tackles the existing challenges and achieves precise ECG data assessment. The transformer model processes ECG data by understanding complex time-dependent relationships which allows it to perform detailed pattern detection for various ECG forms. Extensive experimentation with the model on a full ECG dataset showed an accuracy rate of 99.06% along with 97.93% is precision, 96.69% is recall, 97.31% is F1 score and the Dice coefficient also at 97.31%. The model achieves better than existing state-of-the-art benchmark results regarding CRNN, 1DResNet, ResNetBiGRU which were used for reference assessment. The superior results from the Proposed SkipCNN based ECG Transformer Spike Model show great promise to transform ECG analysis in clinical environments which require fast and precise healthcare solutions. The proposed model promises clinical success in real-time ECG assessments because it demonstrates better accuracy and data noise resilience while functioning across ECG varieties which leads to improved clinical support for medical professionals and enhanced patient treatment results. © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2025.