Song, Insu, Huang, Yi, Koh, Tieh Hee Hai, and Rajadurai, Victor Samuel (2021) Pervasive Monitoring of Gastrointestinal Health of Newborn Babies. In: Lecture Notes in Computer Science LNAI (13031) pp. 359-369. From: PRICAI 2021: 18th Pacific Rim International Conference on Artificial Intelligence, 8-12 November 2021, Hanoi, Vietnam.

PDF (Published Version) - Published Version Restricted to Repository staff only

 

1

Abstract

Babies needing intensive care are at risk of developing gastrointestinal problems, such as feed intolerance and necrotizing enterocolitis (NEC). Monitoring, early detection, and prevention of bowel diseases in newborn may improve outcomes. However, continuous monitoring of the gastrointestinal health of babies is not currently available. We develop an innovative miniature Bowel Sounds Sensor (BoSS) for term babies and a bowel sound analyzer, called Recurrent Local Relation Encoder Classifier (ReLATEC), for real-time, visual monitoring of bowel functions in NICUs. ReLATEC detects types and locations of bowel sounds from a continuous audio stream of bowel activities recorded in noisy hospital environments. ReLATEC combines the advantages of CNN and RNN by using local attention with recurrent layers. We collected 171 bowel sound recordings from 113 newborn babies at two NICUs to evaluate our approach. The bowel sound detector was then trained using weak labels. The detector performed 7% better than conventional approaches. It was shown a sensitivity of 91% and specificity of 71% in detecting short burst bowel sounds. It showed a sensitivity of 97% and specificity of 72% in detecting long burst bowel sounds. Despite the model being trained with weak labels, it detected the boundaries of the two bowel sounds reliably for real-time visual monitoring.

Actions (Repository Staff Only)

Item Control Page