Spatio-Temporal Fuzzy Coding of Video Motion Time Series for Late Onset Sepsis Detection in Neonatalogy.

Area of Science:

• Neonatal Medicine
• Biomedical Engineering
• Data Science

Background:

• Late-onset sepsis is a major cause of mortality in very premature infants (10-25% incidence).
• Current diagnostic methods can be invasive and may not allow for early detection.

Purpose of the Study:

• To develop and validate a non-invasive, video-based method for early detection of late-onset sepsis in very premature newborns.
• To assess the efficacy of motion analysis in differentiating between septic and healthy premature infants.

Main Methods:

• Extraction of 14 distinct motion features from over 2636 hours of video recordings of 16 very premature newborns (8 sepsis, 8 control).
• Processing of motion features using fuzzy spatio-temporal coding and smoothed principal component analysis to generate 16 unique newborn trajectories.
• Comparison of trajectory parameters (position, speed, acceleration) between sepsis and control groups.

Main Results:

• Significant differences in 8 trajectory parameters were identified between newborns with sepsis and control newborns.
• Unsupervised classification based on these parameters achieved an accuracy of 81.3% in distinguishing between infected and healthy infants.
• The method demonstrated potential for early sepsis detection through non-invasive video analysis.

Conclusions:

• Video-based motion analysis is a promising non-invasive tool for early sepsis detection in neonatal intensive care units.
• This approach can facilitate timely interventions, potentially improving outcomes for very premature infants.
• Further research can refine this method for broader clinical application in neonatal care.