Related Experiment Videos
Medical irregular multivariate time series forecasting based on multi-scale temporal-frequency domain patch fusion
Xueping Liu1, Tianyi Gong1, Youru Li2
1College of Artificial Intelligence, Shenyang Aerospace University, Shenyang, China.
Frontiers in Physiology
|May 1, 2026
Summary
Forecasting irregular medical time series is challenging. A new method, Multi-scale Temporal-Frequency domain fusion Patching and Dynamic Graph modeling (MTFP-DG), effectively captures complex dependencies for improved accuracy.
Area of Science:
- Biomedical Informatics
- Data Science
- Time Series Analysis
Background:
- Accurate forecasting of medical multivariate time series is crucial for healthcare monitoring and decision support.
- Physiological data present challenges due to irregular sampling, missing values, and complex temporal/inter-variable dependencies.
Purpose of the Study:
- To propose a novel method, MTFP-DG, for accurate forecasting of irregular multivariate medical time series.
- To address challenges of irregularity, asynchrony, and complex dependencies in physiological data.
Main Methods:
- MTFP-DG transforms irregular time series into multi-scale patches for temporal alignment without interpolation.
- A dual-domain encoding fuses temporal (Transformable Time-aware Convolution Network) and frequency (Irregular Fourier Analysis Network) features.
- Dynamic graphs are constructed to capture evolving inter-variable correlations.
Main Results:
- MTFP-DG demonstrated superior performance over state-of-the-art methods on five real-world medical datasets.
- The method achieved high accuracy in retrospective irregular multivariate time series forecasting benchmarks.
Conclusions:
- Integrating multi-scale patching and dynamic graph modeling effectively captures temporal dependencies and inter-series relationships.
- MTFP-DG offers a robust tool for proactive healthcare planning, pending further clinical validation.
