Complex networks: application to pathology detection in voice signals

Area of Science:

• Complex Systems
• Biomedical Signal Processing
• Network Science

Background:

• Time series analysis often employs non-linear statistics.
• Complex network methodology offers a novel perspective for time series analysis.
• Application of complex networks to pseudo-periodic signals, like voice, is underexplored.

Purpose of the Study:

• To apply complex network techniques to voice signals.
• To differentiate between healthy and pathological voice signals.
• To evaluate the efficacy of complex networks against traditional non-linear statistics for voice analysis.

Main Methods:

• Voice signals were analyzed using the complex network-based technique.
• Statistical measures were computed from the constructed complex networks.
• Lempel-Ziv complexity and sample entropy were used as comparative non-linear statistics.

Main Results:

• Complex network statistics successfully differentiated between healthy and pathological voice signals.
• Traditional non-linear statistics (Lempel-Ziv complexity, sample entropy) did not achieve similar differentiation.
• The complex network approach revealed unique information within voice time series.

Conclusions:

• Complex network analysis provides a powerful tool for voice signal analysis.
• This method offers potential for improved diagnosis of voice pathologies.
• Network-based analysis of voice signals can uncover diagnostic insights missed by other methods.