Contactless video-based assessment of infant non-nutritive sucking using artificial intelligence: a perspective for speech-language pathologists

Non-nutritive sucking (NNS; sucking without nutrition delivery) is one of the earliest motor behaviors in infancy and has key connections to feeding. In research, NNS is typically assessed using contact-based devices that rely on sensorized pacifiers attached to pressure transducers. These methods may disrupt the infant’s natural sucking pattern and are limited to controlled environments. In clinical settings, less objective methods such as use of a gloved finger or observation are common. To overcome these limitations, our team has developed a suite of artificial intelligence (AI) and computer vision-based tools that enable contactless, video-based assessment of NNS. We first established proof-of-concept, comparing facial landmark-derived jaw movement signals from video recordings to data obtained from a contact-based NNS device. We created the first annotated infant facial landmark dataset (InfAnFace), improving the accuracy of a deep learning-based inference model across varied infant appearances and conditions. Then, we developed an end-to-end system capable of detecting and segmenting NNS behavior throughout extended video recordings. These advances have broad practical relevance for in-person and remote assessment of NNS, including in real-world settings. This technology has the potential to positively impact speech-language pathology clinical practice by supporting practical and objective assessment of NNS.