Self-Supervised Contrastive Learning for Medical Time Series: A Systematic Review

Area of Science:

• Medical informatics
• Machine learning
• Time series analysis

Background:

• Medical time series data (e.g., EEG, ECG, ICU readings) offer valuable insights for diagnosis and treatment.
• Labeling medical time series is labor-intensive, time-consuming, and requires expert knowledge, limiting data mining.
• Self-supervised contrastive learning (SCL) has emerged as a powerful technique to learn from unlabeled data.

Conclusions:

• SCL shows significant potential for overcoming annotation limitations in medical time series research.
• Future directions include effective augmentation design, hierarchical time series analysis, and multimodal data processing.
• Early-stage SCL demonstrates promise for advancing medical time series analysis and applications.