Detection of Parkinson's disease using nocturnal breathing signals based on multifractal detrended fluctuation analysis.

Abstract

Parkinson's disease (PD) is a highly prevalent neurodegenerative disorder that poses a significant challenge in terms of accurate and cost-effective diagnosis. This study focuses on the use of fractal features derived from nocturnal breathing signals to diagnose PD. Our study includes 49 individuals with Parkinson's disease (PD group), 49 relatively healthy individuals without PD (HC group), 49 individuals without PD but with other diseases (NoPD group), as well as 12 additional PD patients and 200 healthy individuals for testing. Using multifractal detrended fluctuation analysis, we extracted fractal features from nocturnal breathing signals, with logistic regression models applied to diagnose PD, as demonstrated in receiver operating characteristic curves. Eight fractal features show significant diagnostic potential for PD, including generalized Hurst exponents for the Airflow, Thorax, and Abdomen signals and the multifractal spectrum width of the SaO2 signal. Finally, the area under the receiver operating characteristic curve (AUC) of the training data set of the PD and HC groups for all four signals is 0.911, and the AUC of the testing data set is 0.929. These results demonstrate the potential of this work to enhance the accuracy of PD diagnosis in clinical settings.