172 - Early prediction of brain injuries in term infants based on cerebral near-infrared spectroscopy

Friday, April 28, 2023

5:15 PM – 7:15 PM ET

Poster Number: 172
Publication Number: 172.136

Minoo Ashoori, University College Cork, Cork, Cork, Ireland; Aisling A. Garvey, INFANT Centre, Ireland/Brigham and Womens Hospital, USA, Cork, Cork, Ireland; John M.. O'Toole, INFANT Research Centre, University College Cork, Ireland, Cork, Cork, Ireland; Fiona McDonald, University College Cork, Cork, Cork, Ireland; Ken D. O'Halloran, University College Cork, Cork, Cork, Ireland; Eugene Dempsey, university college cork, Cork, Cork, Ireland

Presenting Author(s)

Minoo Ashoori, MS (she/her/hers)

PhD student
University College Cork
Cork, Cork, Ireland

Background: Hypoxic ischaemic encephalopathy (HIE) is the most common cause of acquired brain injury in term infants, often leading to adverse neurodevelopmental outcome or death. HIE is characterized by abnormal cerebral blood flow (CBF) and metabolism. Cerebral near-infrared spectroscopy (NIRS) can provide real-time information about changes in brain haemodynamics, oxygenation, and metabolism in this population.

Objective: This study examined the potential clinical utility of cerebral NIRS components to detect early adverse outcome using machine learning methods.

Design/Methods: We analyzed 58 infants with all grades of HIE ( >36 weeks GA) enrolled in the Multimodal Assessment of Newborns at Risk of Neonatal Hypoxic Ischaemic Encephalopathy (Monitor) trial. All infants had continuous cerebral oxygenation monitoring for up to 84 hours after birth. Magnetic resonance imaging (MRI) was performed during the first week of life. The NIRS signals were pre-processed, and quantitative features were extracted. Machine learning models were developed to detect abnormal outcome (brain injury on MRI and/or death in the first week) using the leave-one-out cross validation approach.

Results:

The analysis showed that the quantitative features of cerebral NIRS predicted abnormal outcome with the area under the receiver operating characteristic curve (AUC) of 0.699 (95% CI: 0.551 - 0.824), outperforming the models predicting abnormal outcome based on HIE severity (AUC: 0.662, 95% CI: 0.505 - 0.804) and hypothermia treatment (AUC: 0.608, 95% CI: 0.475 - 0.743). The performance of the cerebral NIRS model improved when the prolonged relative desaturations (PRDs; data-driven desaturations lasting 2-15 minutes) were removed, AUC: 0.722, 95% CI: 0.579 - 0.849. Furthermore, we found a significant association between the features extracted from cerebral NIRS and HIE severity (mild, moderate, and severe) with AUC of 0.762 (95% CI: 0.628 – 0.893). The association improved when PRDs were removed from the cerebral NIRS signals (AUC: 0.811, 95% CI: 0.710 - 0.925).

Conclusion(s):

Machine-learning methods provide a scalable, automated analysis of the cerebral NIRS signal and may allow for early objective identification of infants at risk of adverse short-term outcome and aid in grading the severity of HIE.