Newborn Cry-Based Diagnostic System to Distinguish between Sepsis and Respiratory Distress Syndrome Using Combined...

Area of Science:

• Neonatal health diagnostics
• Bioacoustics
• Machine Learning applications

Background:

• Newborn cries are vital communication and health indicators.
• Cry signal analysis is an established biomarker for infant pathologies.
• Discriminating between specific pathology groups using cry signals is an emerging area.

Purpose of the Study:

• To differentiate between newborns with sepsis and Neonatal Respiratory Distress Syndrome (RDS) using cry signal analysis.
• To evaluate the effectiveness of Machine Learning (ML) classifiers, specifically Multilayer Perceptron (MLP) and Support Vector Machine (SVM).
• To explore the utility of fusing spectral (Harmonic Ratio) and short-term (Gammatone Frequency Cepstral Coefficients) features.

Main Methods:

• Analysis of cry signals from two perspectives: musical (spectral features) and speech processing (short-term features).
• Implementation and fine-tuning of MLP and SVM machine learning models.
• Fusion of Harmonic Ratio (HR) and Gammatone Frequency Cepstral Coefficients (GFCCs) into a combined feature set.

Main Results:

• The combined feature set, normalized and fused, significantly improved classification performance.
• Support Vector Machine (SVM) achieved the highest accuracy at 95.3%, outperforming Multilayer Perceptron (MLP).
• The study demonstrated the effectiveness of combining diverse feature modalities for enhanced cry signal analysis.

Conclusions:

• Fusing features from spectral and short-term analyses provides a more powerful approach to cry signal interpretation.
• Machine learning models, particularly SVM, show significant potential for distinguishing complex neonatal pathologies like sepsis and RDS.
• The findings support further research with larger datasets and additional pathological conditions.