DISCOVERING CAUSALITIES FROM CARDIOTOCOGRAPHY SIGNALS USING IMPROVED CONVERGENT CROSS MAPPING WITH GAUSSIAN PROCESSES

Area of Science:

• Time series analysis
• Causal inference
• Non-parametric Bayesian methods

Background:

• Convergent cross mapping (CCM) is a method for causal discovery in coupled time series.
• Standard CCM is sensitive to observation noise and requires parameter tuning via grid search.
• Granger causality has limitations due to its separability assumption.

Purpose of the Study:

• To develop an improved, noise-robust version of CCM for causal discovery in time series.
• To apply Gaussian processes within a Bayesian framework to enhance CCM.
• To investigate the causality between fetal heart rate and uterine activity.

Main Methods:

• Utilized Gaussian processes for state space reconstruction and causal discovery.
• Implemented a non-parametric Bayesian probabilistic framework for CCM.
• Validated the novel approach on simulated data and real-world obstetrics data.

Main Results:

• The proposed Gaussian process-based CCM effectively handles noisy time series data.
• Demonstrated successful causal discovery in simulated datasets.
• Identified a causal link from uterine activity to fetal heart rate in late pregnancy.

Conclusions:

• Gaussian processes offer a principled and robust enhancement to CCM for noisy time series.
• The findings support the hypothesis that uterine activity influences fetal heart rate.
• This method has potential applications in analyzing complex biomedical signals.