Cyclostationary phase conditioning improves medical time series restoration
Cyclostationary Phase Conditioning for Medical Time Series Diffusion
Machine LearningArtificial Intelligence
Summary
Some medical signals like heartbeats and brain waves follow repeating cycles, but noise and interference make it hard to spot important details for diagnosis. The authors show that explicitly using the signals' repeating cycle patterns during restoration helps recover clearer signals. They introduced special ways to represent these cycles and measure when this method will help most. Their approach also reduces the computing needed to clean the signals while keeping good results.
What this means in practice
- •For medical device engineers: Improve wearable heart and brain monitors by restoring clearer physiological signals obscured by noise.$Commercial implications: Enables clearer physiological signal recovery in medical devices, enhancing diagnostic accuracy and device market value.
- •For biomedical data scientists: Develop better software tools to clean noisy cardiac and brain time series data using explicit cyclic phase information.
Authors
Samuel Ruiperez-Campillo, Michele Copetti, Jorge da Silva Goncalves, Sonia Laguna, Julia E. Vogt
Abstract
Many physiological time series, such as cardiac and brain recordings, exhibit cyclostationarity: their statistics vary periodically with an underlying cycle phase. Corruption from motion, poor contact, and physiological interference obscures morphology needed for diagnosis, making signal restoration essential. Existing diffusion approaches condition on corrupted observations alone and must learn cyclic structure implicitly. We instead propose two inductive biases which encode cyclostationarity: a shift-covariant wavelet representation and dense per-sample phase conditioning inferred from the corrupted input. We further introduce a training-free cyclostationarity index that quantifies phase structure and predicts when phase conditioning will help. Finally, we propose antithetic coupling of reverse trajectories to reduce sampling variance while achieving comparable performance with fivefold fewer network evaluations. Across modalities, our results show that explicitly encoding measurable cyclic structure improves physiological time-series restoration.