AI-Based Thermal Video Analysis in Privacy-Preserving Healthcare: A Case Study on Detecting Time of Birth

Jorge García-Torres, Øyvind Meinich-Bache, Siren Rettedal, Kjersti Engan · 2025

Read AI-Based Thermal Video Analysis in Privacy-Preserving Healthcare: A Case Study on Detecting Time of Birth by Jorge García-Torres, Øyvind Meinich-Bache, Siren Rettedal, Kjersti Engan (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2502-04365.txt.

Introduction

Jorge García-Torres and co-authors published a 2025 paper on a system that spots the exact moment of a baby's birth from heat video. About one in ten newborns needs help to start breathing. Care must begin fast, so knowing the precise birth time matters for records and for improving resuscitation. In practice, staff note the time by hand, often only to the nearest minute.

Thermal cameras record heat rather than visible light. Faces and bodies cannot be identified from such footage, which protects the privacy of mothers and staff. The team trained an AI system to find the birth moment in these clips. In tests, it flagged the moment with high precision and caught nearly all true cases. For most test clips it named the time within about a second of the hand-checked value. The authors offer the system as a reliable aid for maternity wards.

Catalog record

CREATORJorge García-Torres, Øyvind Meinich-Bache, Siren Rettedal, Kjersti Engan
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE22 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2502-04365

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

AI-Based Thermal Video Analysis in Privacy-Preserving Healthcare: A Case Study on Detecting Time of Birth Jorge García-Torres, Øyvind Meinich-Bache, Siren Rettedal, Kjersti Engan Abstract Approximately 10% of newborns need some assistance to start breathing and 5% proper ventilation. It is crucial that interventions are initiated as soon as possible after birth. Accurate documentation of Time of Birth (ToB) is thereby essential for documenting and improving newborn resuscitation performance. However, current clinical practices rely on manual recording of ToB, typically with minute precision. In this study, we present an AI-driven, video-based system for automated ToB detection using thermal imaging, designed to preserve the privacy of healthcare providers and mothers by avoiding the use of identifiable visual data. Our approach achieves 91.4% precision and 97.4% recall in detecting ToB within thermal video clips during performance evaluation. Additionally, our system successfully identifies ToB in 96% of test cases with an absolute median deviation of 1 second compared to manual annotations. This method offers a reliable solution for improving ToB documentation and enhancing newborn resuscitation outcomes. † † address: ⋆ Dept. of Electrical Engineering and Computer Science, University of Stavanger, Norway † Laerdal Medical AS, Stavanger, Norway ‡ Dept. for Simulation-based Learning, Stavanger University Hospital, Norway § Faculty of Health Sciences, University of Stavanger, Norway 1 Introduction Figure 1 : Overview of our proposed video-based, AI-driven Time of Birth detector. Birth episodes are recorded by a thermal camera installed on the ceiling. First, thermal videos undergo adaptive normalization using Gaussian Mixture Models, followed by trimming into video clips. At each second, a sliding window containing the previous 25 frames (3 seconds) is utilized as input x(t) to our model for prediction. All the model’s output scores \hat{y}(t) are then concatenated and post-processed using a Finite Impulse Response (FIR) filter with a rectangular window. Finally, Time of Birth \hat{T}_{birth} is inferred based on the model’s confidence in successfully detecting the birth in the video clips. Approximately 10% of newborns need some assistance to start breathing, and 5% proper ventilation, with birth asphyxia as one of the leading causes of neonatal mortality, responsible for an estimated 900,000 deaths annually [ 1 , 2 ] . The Time of Birth (ToB) is defined as the moment when the newborn’s head, torso, and nates are fully visible outside the mother’s perineum, and mark the critical reference point for resuscitation interventions [ 3 ] . According to neonatal resuscitation guidelines [ 4 , 5 ] , Newborn Resuscitation Algorithm Activities (NRAA) should be started within the “golden minute” – the first 60 seconds after birth, as immediate action can significantly reduce the risks of death or long-term complications due to birth asphyxia [ 1 ] . The NRAA guidelines are based on best practices and evidence research needs to be sought [ 6 ] . NRAA timelines are crucial for retrospective analysis and debriefing, research on optimal birth asphyxia treatment, and potential real-time clinical decision support. Accurate ToB documentation is an essential part of an effective NRAA timeline; however, in clinical practice, ToB is recorded manually, often with minute precision. Such lack of precision reduces the reliability of NRAA timeline documentation. Artificial Intelligence (AI) has demonstrated promise in automating NRAA timelines [ 7 , 8 ] , offering a way to streamline the documentation of newborn resuscitation activities. Yet, existing AI-based approaches still depend on manually recorded ToB, limiting the broader applicability of these systems in critical care settings. The NewbornTime project [ 9 ] seeks to address these gaps by developing a fully automated system that generates accurate NRAA timelines, including ToB and resuscitation activities. One of the key innovations is our privacy-preserving method, which utilizes thermal imaging to detect ToB by identifying the higher skin temperature of newborns compared to other persons in the room. Thermal or infrared (IR) imaging captures the heat emissions, forming images that show temperature differences. This technique ensures the privacy of both healthcare providers and mothers by avoiding identifiable visual data. In healthcare, such technologies can be pivotal in building trust and maintaining compliance with data protection regulations, such as the General Data Protection Regulation (GDPR). In [ 10 ] we demonstrated that AI-driven, image-based systems using individual thermal frames had potential for estimating ToB. However, the lack of temporal context led to imprecision in certain scenarios. To the author’s knowledge, no other work on automatic ToB detection exists. The primary contribution of this work is a spatiotemporal AI-based method for ToB detection using privacy-preserving thermal video. By analyzing continuous changes in movement and thermal characteristics, the model captures the dynamic birth process, resulting in a more precise ToB documentation. 2 Data Material The dataset contains a total of 321 birth videos captured by thermal cameras. Approximately 80% of the births occurred in supine position, 15% in side-lying position, and 5% in hands-and-knees position. Data were collected in a semiautomatic manner [ 11 ] at Stavanger University Hospital (SUS), Norway, using a passive thermal module supplied by Mobotix [ 12 ] and installed in eight delivery rooms. The sensor was mounted to the ceiling, centered above the head of the mother. The recording was triggered when any pixel’s temperature exceeded 30 ∘ C, detecting human presence and streamlining data capture. ToB was manually registered with second precision by midwives or nurse assistants using the Liveborn Observation App [ 13 ] , specifically designed to document post-birth events in research projects, providing more accurate ToB logging than standard practice [ 14 ] . Pressing the “Baby Born” button in this app automatically saved a 30-minute thermal video generated at 252 \times 336 resolution and 8.33 fps, covering 15 minutes before and after birth. The rest of the video was discarded. The registration of ToB in Liveborn was sometimes delayed due to all the activities that happened around the birth. Therefore, in this study, manual annotation of the ToB with second precision was done by carefully inspecting the thermal videos. 3 Methods In this work, we introduce an AI-driven, video-based system that uses spatiotemporal information to estimate ToB. The task is formulated as a binary classification problem where we want to identify the birth within video clips. We propose a pipeline where thermal videos are first normalized with an adaptive method based on Gaussian Mixture Models (GMM) and then streamed into our model using a sliding window with a constant stride for generating predictions. An overview of the proposed method is illustrated in Figure 1 . Going forward, let \mathcal{V}\in\mathbb{R}^{N\times H\times W} represent a single-channel thermal video with N thermal frames and a spatial resolution of H\times W . I(n) denote the thermal frame at index n=0,1,...,N-1 . We define a video clip V(n) as the preceding F frames: V(n)=\{I(n),I(n-1),...,I(n-F+1)\} (1) 3.1 Video preprocessing In García-Torres et al. [ 15 ] , we explored the challenges of using thermal sensors for ToB detection, finding limitations in relying on absolute temperature values due to several distorting factors such as autocalibration and room temperature. This potential reliability issue required the use of relative temperatures and introduced a normalization challenge for our AI system. To address this, we later proposed a GMM-based normalization method [ 10 ] to effectively identify a relevant temperature range based on human

Read the complete text