Advancing newborn care: Precise time of birth detection using ai-driven thermal imaging with adaptive normalization

Area of Science:

• Medical Technology
• Artificial Intelligence
• Neonatal Care

Background:

• 5-10% of newborns require breathing assistance, highlighting the need for improved resuscitation practices.
• Current methods for recording Time of Birth (ToB) lack precision and objectivity, hindering analysis of neonatal resuscitation events.
• Accurate ToB is crucial for timely interventions within the critical "golden minute" after birth.

Purpose of the Study:

• To develop the first Artificial Intelligence (AI)-driven detector for precise Time of Birth (ToB) estimation using thermal imaging.
• To address the limitations of manual ToB recording, which is often imprecise and prone to errors.
• To enhance newborn resuscitation care through objective data collection and analysis.

Main Methods:

• Utilized a three-step methodology involving AI and thermal imaging.
• Developed an adaptive normalization method using Gaussian mixture models (GMM) to handle thermal sensor inconsistencies.
• Implemented an AI model for newborn detection in thermal frames and post-processing for ToB estimation.

Main Results:

• Achieved 88.1% precision and 89.3% recall in detecting newborns within thermal video frames.
• Estimated Time of Birth (ToB) with an absolute median deviation of 2.7 seconds compared to manual annotations.
• Demonstrated the potential of thermal imaging and AI for accurate and private newborn detection.

Conclusions:

• The AI-driven thermal imaging approach offers a promising, privacy-preserving method for accurate Time of Birth (ToB) determination.
• This technology can significantly improve the analysis of neonatal resuscitation events and enhance clinical decision-making.
• The developed system addresses critical gaps in objective data collection for neonatal emergency care.