Frequency Domain mixup Augmentation and logit Compensation for Self-Supervised Multi-label Imbalanced Electrocardiogram Classification

doi:10.3778/j.issn.1673-9418.2405065

Abstract

Abstract: Self-supervised contrastive learning has been proven effective in learning good feature representations by contrasting views through data augmentation, followed by fine-tuning for downstream (classification) tasks. Thus, it obtains wide applications. Electrocardiogram (ECG), as a non-invasive, low-risk, and low-cost signal source for cardiovascular diseases, its classification aids in early prevention and precise treatment of conditions like arrhythmia. However, most existing methods for ECG representation learning only perform contrastive learning through temporal perturbation augmentation of examples, overlooking the potential utilization of frequency-domain information, leaving room for further improvement in representation quality. Therefore, a frequency domain mixup augmentation strategy is designed for ECG samples, which generates augmented samples by exchanging frequency domain information between samples to achieve contrastive learning, thus addressing the shortcomings of existing ECG representation learning. In the downstream fine-tuning stage, considering that ECG classification inherently involves a multi-label class imbalance problem, this paper proposes mitigating this issue by incorporating label frequencies into binary cross-entropy (BCE) loss as logit compensation. Finally, model evaluation is conducted on the CPSC2018 and Chapman datasets. Experimental results demonstrate that integrating the proposed method as an independent module into multiple baseline models improves performance in terms of AUC and mAP metrics. Particularly, significant enhancements are observed in the performance of certain rare disease indicators, thereby validating the effectiveness of this approach.

Key words: electrocardiogram classification, arrhythmia, self-supervised contrastive learning, multi-label, class imbalance

摘要： 自监督对比学习通过数据增广视图间的对比已被证明能习得好的特征表征，继而通过微调完成下游（分类）任务，因此得到广泛应用。心电图（ECG）作为非侵入、低风险和低成本的心血管疾病常用信号源，其分类有助于早期预防和精确治疗心率失常等。然而现有针对ECG表征学习的大多数方法仅通过对样本进行时域的扰动增广进行对比学习，其忽略了频域潜在的信息利用，留下了进一步提升表征质量的空间。为此，针对ECG样本设计了一个频域mixup的增广策略，通过交换样本间的频域信息生成原始样本的增广实现对比学习，弥补了现有ECG表征学习的不足。在下游微调阶段，考虑到ECG分类本质上属于多标记的类不平衡问题，提出了结合标签频率对二元交叉熵（BCE）损失作logit补偿缓和该问题。最后在CPSC2018和Chapman数据集上进行模型评估，实验结果表明提出的方法作为独立模块插入至多个基线模型在AUC和mAP指标上均有提高，尤其是个别罕见疾病性能指标提升显著，从而验证了该方法的有效性。

关键词: 心电图分类, 心率失常, 自监督对比学习, 多标记, 类不平衡

CAO Siyuan, CHEN Songcan. Frequency Domain mixup Augmentation and logit Compensation for Self-Supervised Multi-label Imbalanced Electrocardiogram Classification[J]. Journal of Frontiers of Computer Science and Technology, 2025, 19(4): 1011-1020.

操思源, 陈松灿. 频域mixup增广和logit补偿的自监督多标记不平衡心电图分类[J]. 计算机科学与探索, 2025, 19(4): 1011-1020.

References

[1] REHMAN S, REHMAN E, IKRAM M, et al. Cardiovascular disease (CVD): assessment, prediction and policy implications[J]. BMC Public Health, 2021, 21(1): 1299.
[2] MELZI P, TOLOSANA R, CECCONI A, et al. Analyzing artificial intelligence systems for the prediction of atrial fibrillation from sinus-rhythm ECGs including demographics and feature visualization[J]. Scientific Reports, 2021, 11(1): 22786.
[3] YAN G S, LIANG S, ZHANG Y C, et al. Fusing transformer model with temporal features for ECG heartbeat classification[C]//Proceedings of the 2019 IEEE International Conference on Bioinformatics and Biomedicine. Piscataway:IEEE, 2019: 898-905.
[4] 李兴秀, 唐建军, 华晶. 结合CNN与双向LSTM的心律失常分类[J]. 计算机科学与探索, 2021, 15(12): 2353-2361.
LI X X, TANG J J, HUA J. Arrhythmia classification based on CNN and bidirectional LSTM[J]. Journal of Frontiers of Computer Science and Technology, 2021, 15(12): 2353-2361.
[5] LI F, LI H X, LI Y, et al. Decoupling representation learning for imbalanced electroencephalography classification in rapid serial visual presentation task[J]. Journal of Neural Engineer-ing, 2022, 19(3): 036011.
[6] KANG B Y, XIE S N, ROHRBACH M, et al. Decoupling representation and classifier for long-tailed recognition[EB/OL]. [2024-02-18]. https://arxiv.org/abs/1910.09217.
[7] YANG Y Z, XU Z. Rethinking the value of labels for improving class-imbalanced learning[EB/OL]. [2024-02-18]. https:// arxiv.org/abs/2006.07529.
[8] ZHANG C, LIU H, LI J, et al. Contrastive graph few-shot learning[EB/OL]. [2024-02-18]. https://arxiv.org/abs/2210.00084.
[9] RADFORD A, WU J, CHILD R, et al. Language models are unsupervised multitask learners[J]. OpenAI Blog, 2019, 1(8): 9.
[10] CHEN T, KORNBLITH S, NOROUZI M, et al. A?simple framework?for?contrastive?learning?of?visual?representations[C]//Proceedings of the 37th International Conference on Machine Learning, 2020:?1597-1607.
[11] MEHARI T, STRODTHOFF N. Self-supervised representation learning from 12-lead ECG data[J]. Computers in Biology and Medicine, 2022, 141: 105114.
[12] KIYASSEH D, ZHU T, CLIFTON D. CLOCS: contrastive learning of cardiac signals[C]//Proceedings of the 38th International Conference on Machine Learning, 2021: 5606-5615.
[13] WANG Y H, HAN Y, WANG H S, et al. Contrast everything: a hierarchical contrastive framework for medical time-series[EB/OL]. [2024-03-14]. https://arxiv.org/abs/2310.14017.
[14] 韩涵, 黄训华, 常慧慧, 等. 心电领域中的自监督学习方法综述[J]. 计算机科学与探索, 2024, 18(7): 1683-1704.
HAN H, HUANG X H, CHANG H H, et al. Review of self-supervised learning methods in field of ECG[J]. Journal of Frontiers of Computer Science and Technology, 2024, 18(7): 1683-1704.
[15] LIU Z Y, ALAVI A, LI M Y, et al. Self-supervised contrastive learning for medical time series: a systematic review[J]. Sensors, 2023, 23(9): 4221.
[16] ZHAO R, YAN R Q, CHEN Z H, et al. Deep learning and its applications to machine health monitoring[J]. Mechanical Systems and Signal Processing, 2019, 115: 213-237.
[17] PENG F R, LUO J C, LU X, et al. Cross-domain contrastive learning for time series clustering[C]//Proceedings of the 38th AAAI Conference on Artificial Intelligence and the 36th Conference on Innovative Applications of Artificial Intelligence and the 14th Symposium on Educational Advances in Artificial Intelligence. New York: ACM, 2024: 8921-8929.
[18] GUO H, WANG S. Long-tailed multi-label visual recognition by collaborative training on uniform and re-balanced samplings[C]//Proceedings of the 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2021: 15089-15098.
[19] PHAN T, LE D, BRIJESH P, et al. Multimodality multi-lead ECG arrhythmia classification using self-supervised learning[C]//Proceedings of the 2022 IEEE-EMBS International Conference on Biomedical and Health Informatics. Piscataway: IEEE, 2022: 1-4.
[20] P?PPELBAUM J, CHADHA G S, SCHWUNG A. Contrastive learning based self-supervised time-series analysis[J]. Applied Soft Computing, 2022, 117: 108397.
[21] SOLTANIEH S, ETEMAD A, HASHEMI J. Analysis of augmentations for contrastive ECG representation learning[C]//Proceedings of the 2022 International Joint Conference on Neural Networks. Piscataway: IEEE, 2022: 1-10.
[22] CHARTE F, RIVERA A J, DEL JESUS M J, et al. MLSMOTE: approaching imbalanced multilabel learning through synthetic instance generation[J]. Knowledge-Based Systems, 2015, 89: 385-397.
[23] CHAWLA N V, BOWYER K W, HALL L O, et al. SMOTE: synthetic minority over-sampling technique[J]. Journal of Artificial Intelligence Research, 2002, 16(1): 321-357.
[24] CHARTE F, RIVERA A, DEL JESUS M J, et al. A first approach to deal with imbalance in multi-label datasets[M]//Hybrid artificial intelligent systems. Berlin, Heidelberg: Springer, 2013: 150-160.
[25] CHEN K, LU B L, KWOK J T. Efficient classification of multi-label and imbalanced data using min-max modular classifiers[C]//Proceedings of the 2006 IEEE International Joint Conference on Neural Network. Piscataway: IEEE, 2006: 1770-1775.
[26] TAHIR M A, KITTLER J, BOURIDANE A. Multilabel classification using heterogeneous ensemble of multi-label classifiers[J]. Pattern Recognition Letters, 2012, 33(5): 513-523.
[27] LIU Y, LI Q C, WANG K Q, et al. Automatic multi-label ECG classification with category imbalance and cost-sensitive thre-sholding[J]. Biosensors, 2021, 11(11): 453.
[28] LIU J X, CHEN S C. TimesURL: self-supervised contrastive learning for universal time series representation learning[J]. Proceedings of the AAAI Conference on Artificial Intelligence, 2024, 38(12): 13918-13926.
[29] CHENG J Y, GOH H, DOGRUSOZ K, et al. Subject-aware contrastive learning for biosignals[EB/OL]. [2024-03-14]. https://arxiv.org/abs/2007.04871.
[30] CAO K D, WEI C, GAIDON A, et al. Learning imbalanced datasets with label-distribution-aware margin loss[C]//Advances in Neural Information Processing Systems 32, 2019: 1567-1578.
[31] MENON A, JAYASUMANA S, RAWAT A, et al. Long-tail learning via logit adjustment[EB/OL]. [2024-03-14]. https://arxiv.org/abs/2007.07314.
[32] TAN J, WANG C, LI B, et al. Equalization loss for long-tailed object recognition[C]//Proceedings of the 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2020: 11662-11671.
[33] XIE Y T, QIN L, TAN H C, et al. Automatic 12-leading electro-cardiogram classification network with deformable convolu-tion[C]//Proceedings of the 2021 43rd Annual International Conference of the IEEE Engineering in Medicine & Biology Society. Piscataway: IEEE, 2021: 882-885.
[34] REYNA M A, ALDAY E A P, GU A N, et al. Classification of 12-lead ECGs: the PhysioNet/computing in cardiology challenge 2020[C]//Proceedings of the 2020 Computing in Cardiology. Piscataway: IEEE, 2020: 1-4.
[35] STRODTHOFF N, WAGNER P, SCHAEFFTER T, et al. Deep learning for ECG analysis: benchmarks and insights from PTB-XL[J]. IEEE Journal of Biomedical and Health Informatics, 2021, 25(5): 1519-1528.
[36] SCHAPIRE R E, SINGER Y. BoosTexter: a boosting-based system for text categorization[J]. Machine Learning, 2000, 39(2): 135-168.
[37] BAI S J, KOLTER J Z, KOLTUN V, et al. An empirical evaluation of generic convolutional and recurrent networks for sequence modeling[EB/OL]. [2024-03-14]. https://arxiv.org/abs/1803.01271.
[38] PASZKE A, GROSS S, MASSA F, et al. An imperative style, high-performance deep learning library[C]//Advances in Neural Information Processing Systems 32, 2019: 8024-8035.
[39] MAHARANA K, MONDAL S, NEMADE B. A review: data pre-processing and data augmentation techniques[J]. Global Transitions Proceedings, 2022, 3(1): 91-99.
[40] GARCEA F, SERRA A, LAMBERTI F, et al. Data augmentation for medical imaging: a systematic literature review[J]. Computers in Biology and Medicine, 2023, 152: 106391.
[41] ZHANG W R, GENG S J, HONG S D. A simple self-supervised ECG representation learning method via manipulated temporal-spatial reverse detection[J]. Biomedical Signal Pro-cessing and Control, 2023, 79: 104194.
[42] SARKAR P, LOBMAIER S, FABRE B, et al. Detection of maternal and fetal stress from the electrocardiogram with self-supervised representation learning[J]. Scientific Reports, 2021, 11(1): 24146.
[43] SARKAR P, ETEMAD A. Self-supervised ECG representation learning for emotion recognition[J]. IEEE Transactions on Affective Computing, 2020, 13(3): 1541-1554.