INFERENCE ABOUT CAUSALITY FROM CARDIOTOCOGRAPHY SIGNALS USING GAUSSIAN PROCESSES.

Area of Science:

• Time Series Analysis
• Statistical Modeling
• Machine Learning

Background:

• Granger causality is crucial for understanding time series interactions.
• Traditional methods often rely on autoregressive models, which can be sensitive to noise.
• Discovering causal relationships in noisy biomedical signals presents significant challenges.

Purpose of the Study:

• To propose a novel and simple method for discovering Granger causality in noisy time series using Gaussian processes.
• To demonstrate that Granger causality information is encoded within Gaussian process hyperparameters.
• To apply the method to analyze the interaction between fetal heart rate and uterine activity.

Main Methods:

• Utilizing Gaussian processes instead of traditional autoregressive models for Granger causality.
• Extracting Granger causality information from the hyperparameters of the Gaussian processes.
• Validating the proposed method on simulated noisy time series data.

Main Results:

• The proposed Gaussian process method successfully identifies Granger causality in noisy time series.
• Information regarding Granger causality is demonstrably encoded within the hyperparameters.
• Analysis of obstetric data reveals that uterine activity significantly affects fetal heart rate.

Conclusions:

• Gaussian processes offer a robust alternative for Granger causality discovery in noisy data.
• The method provides insights into the complex interactions observed in biomedical time series.
• Findings support the hypothesis that uterine activity impacts fetal heart rate prior to delivery.