Accurate and reliable skin temperature monitoring is critical for the thermoregulation of premature infants, but current methods using wired sensors are invasive and prone to error. Infrared thermography offers a non-invasive, wireless alternative to current wired temperature monitoring methods. This work presents a novel non-contact system for monitoring skin temperature of premature infants in closed incubators using infrared thermography. To enable temperature monitoring of specific body parts, an automated body part segmentation model was developed. The multimodal U-Net incorporated color and long-wave infrared image data. Transfer learning and data augmentation techniques improved the performance of the model, resulting in an average Intersection over Union of 0. 77 across all body parts (head, torso, arms, and legs). To ensure the accuracy of the thermal data itself, a novel temperature correction algorithm was developed. This compensated for systematic errors caused by factors such as an infrared window, reflections from incubator walls, and camera drift. The algorithm achieved a mean absolute error of 0. 17 ^ C and a maximum error of 0. 83 ^ C when validated against a blackbody reference. By combining these steps, our system extracts spatial temperature using a percentile-based method, resulting in a mean absolute error of 0. 41 ^ C compared to a reference adhesive temperature sensor on the torso. The analysis revealed that the torso and arms provided more robust central and peripheral temperature measurements than the head and legs. These results demonstrate the potential of this non-contact system for accurate and reliable clinical temperature monitoring in premature infants, offering significant benefits in terms of patient comfort, reduced risk of infection and improved workflow for medical staff.
Voss et al. (2026) studied this question.