A novel baseline perspective visibility graph for time series analysis

The natural visibility graph method has been widely used in physiological signal analysis, but it fails to accurately handle signals with data points below the baseline. Such signals are common across various physiological measurements, including electroencephalograph (EEG) and functional magnetic resonance imaging (fMRI), and are crucial for insights into physiological phenomena. This study introduces a novel method, the baseline perspective visibility graph (BPVG), which can analyze time series by accurately capturing connectivity across data points both above and below the baseline. We present the BPVG construction process and validate its performance using simulated signals. Results demonstrate that BPVG accurately translates periodic, random, and fractal signals into regular, random, and scale-free networks respectively, exhibiting diverse degree distribution traits. Furthermore, we apply BPVG to classify Alzheimer’s disease (AD) patients from healthy controls using EEG data and identify non-demented adults at varying dementia risk using resting-state fMRI (rs-fMRI) data. Utilizing degree distribution entropy derived from BPVG networks, our results exceed the best accuracy benchmark (77.01%) in EEG analysis, especially at channels F4 (78.46%) and O1 (81.54%). Additionally, our rs-fMRI analysis achieves a statistically significant classification accuracy of 76.74%. These findings highlight the effectiveness of BPVG in distinguishing various time series types and its practical utility in EEG and rs-fMRI analysis for early AD detection and dementia risk assessment. In conclusion, BPVG’s validation across both simulated and real data confirms its capability to capture comprehensive information from time series, irrespective of baseline constraints, providing a novel method for studying neural physiological signals.