Advancing Newborn Care: Precise Birth Time Detection Using AI-Driven Thermal Imaging with Adaptive Normalization

TL;DR

This work tackles the challenge of obtaining precise Time of Birth (ToB) for newborn resuscitation analysis by combining AI with thermal imaging. It proposes a three-step pipeline: adaptive Gaussian Mixture Model normalization to standardize relative temperatures across cameras, a CNN-based frame-level newborn detector using EfficientNetV2B1, and FIR post-processing to estimate ToB from detection scores. The approach achieves a precision of 88.1%, recall of 89.3%, a MCC of 0.866, and a median ToB error of 2.7 seconds, demonstrating potential for automated NRAA timeline generation while preserving privacy. The study advances newborn care analytics by enabling second-precision ToB estimation in real-world multi-camera thermal settings, with implications for training, debriefing, and quality improvement.

Abstract

Around 5-10\% of newborns need assistance to start breathing. Currently, there is a lack of evidence-based research, objective data collection, and opportunities for learning from real newborn resuscitation emergency events. Generating and evaluating automated newborn resuscitation algorithm activity timelines relative to the Time of Birth (ToB) offers a promising opportunity to enhance newborn care practices. Given the importance of prompt resuscitation interventions within the "golden minute" after birth, having an accurate ToB with second precision is essential for effective subsequent analysis of newborn resuscitation episodes. Instead, ToB is generally registered manually, often with minute precision, making the process inefficient and susceptible to error and imprecision. In this work, we explore the fusion of Artificial Intelligence (AI) and thermal imaging to develop the first AI-driven ToB detector. The use of temperature information offers a promising alternative to detect the newborn while respecting the privacy of healthcare providers and mothers. However, the frequent inconsistencies in thermal measurements, especially in a multi-camera setup, make normalization strategies critical. Our methodology involves a three-step process: first, we propose an adaptive normalization method based on Gaussian mixture models (GMM) to mitigate issues related to temperature variations; second, we implement and deploy an AI model to detect the presence of the newborn within the thermal video frames; and third, we evaluate and post-process the model's predictions to estimate the ToB. A precision of 88.1\% and a recall of 89.3\% are reported in the detection of the newborn within thermal frames during performance evaluation. Our approach achieves an absolute median deviation of 2.7 seconds in estimating the ToB relative to the manual annotations.

Figures (12)

• Figure 1: Overview of our proposed AI-driven Time of Birth detector. Birth episodes are captured using a thermal camera installed on the ceiling. Step I: our proposed Gaussian Mixture Model (GMM) normalization is employed in order to overcome the non-consistent distribution in the temperature values produced by other common normalization approaches. Step II: normalized frames are utilized as input $x(n)$ to our model to detect the presence of a newborn. We assess the model's ability to identify the newborn using GradCAM, providing insights into its decision-making process. Step III: the model's output scores $\hat{y}(n)$ are post-processed using a Finite Impulse Response (FIR) filter with a rectangular window. ToB is inferred based on the model's confidence in successfully detecting the newborn in individual video frames.
• Figure 2: Illustration of the thermal module set-up in delivery rooms at SUS. On the left, the thermal sensor is mounted to the ceiling above the head of the mother, marked with a green circle. On the right, there is a representation of the thermal data from the birth scenario, with humans easily detectable due to their body temperature. The newborn, depicted in red, appears slightly warmer than the normal skin temperature at the time of birth. The illustration was created using Adobe Illustrator and published with permission from Laerdal Medical AS, attributed to Fredrik Kleppe.
• Figure 3: Temperature measurements on the surface of a black body device from the thermal sensor modules in a delivery room and the measured ambient room temperature along time. Variations in the measurements range between 35 and 38 degrees Celsius when the black body temperature is expected to be 36$\pm$0.2$^\circ$C. Illustration extracted from garcia2022towards and modified.
• Figure 4: Illustration of raw thermal images. For visualization purposes, temperature values are clipped to a range of 20 to 40 degrees Celsius, representing the typical range of real-world temperatures for objects in the birth scenario. The first row shows the expected thermal visualizations, where human skin temperature is about 35$^\circ$ in delivery rooms and slightly lower in the operation theater. The second row highlights thermal images affected by temperature distortion factors, resulting in objects appearing either warmer (left, with newborn skin temperature exceeding 40$^\circ$) or cooler (right, with human skin temperature just above 30$^\circ$) than the expected temperature value. The third row demonstrates examples of thermal camera miscalibration, where temperature readings are significantly outside the expected range.
• Figure 5: Representation of the complex data distribution of a thermal video and the density estimation provided by a GMM of three Gaussian components using temperature values from a thermal video. In blue, orange, and green, the three GMM components extracted from the data distribution. In red, the cumulative sum of all the GMM components.