2026-09-20T21:44:02Zhttps://cora.ucc.ie/server/oai/request

oai:cora.ucc.ie:10468/170592025-02-19T02:03:39Zcom_10468_386com_10468_4com_10468_1093com_10468_1col_10468_387col_10468_1094col_10468_8983

Chapter 4 for 1 year McDonald, Fiona O'Halloran, Ken Ashoori, Minoo Science Foundation Ireland 2025-02-18T11:18:20Z 2025-02-18T11:18:20Z 2024 2024 Preterm and term infants are vulnerable to brain injuries resulting from inadequate cerebral oxygenation, posing significant risks to their long-term neurodevelopment. Near-infrared spectroscopy (NIRS) is a non-invasive technology capable of monitoring regional cerebral oxygen saturation (rcSO2) and detecting cerebral desaturation events in neonates. However, the clinical utility of NIRS signals remains limited due to challenges in interpreting complex patterns. This thesis aimed to enhance the diagnostic and prognostic value of NIRS by employing advanced signal processing and artificial intelligence (AI) techniques. We developed and optimized methods to extract prolonged relative desaturations (PRDs) from NIRS signals and combined these with machine learning (ML) and deep learning approaches. These models integrated quantitative features derived from rcSO2 and peripheral oxygen saturation (SpO2) signals to predict brain injuries, such as intraventricular hemorrhage (IVH) in preterm infants and hypoxic-ischemic encephalopathy (HIE) in term infants. Our findings demonstrated that features extracted from PRDs significantly outperformed traditional threshold-based approaches in predicting adverse outcomes, achieving an area under the receiver operating characteristic curve (AUC) of 0.846 for IVH detection in preterm infants. Additionally, machine learning models revealed a significant association between rcSO2 patterns and adverse outcomed assessed by MRI, with AUC values reaching 0.73. Deep learning methods further automated feature extraction, providing modest accuracy in detecting MRI-confirmed brain injuries. These results highlight the potential of integrating NIRS with AI-driven analysis to improve early detection and management of neonatal brain injuries. This research lays the foundation for more personalized and timely interventions, ultimately advancing neonatal care and outcomes. Not peer reviewed Accepted Version application/pdf Ashoori, M. 2024. AI-based analysis of cerebral oxygenation in preterm and term infants. PhD Thesis, University College Cork. 321 https://hdl.handle.net/10468/17059 en University College Cork info:eu-repo/grantAgreement/SFI/SFI Starting Investigator Research Grant/15/SIRG/3580/IE/Advancing Neuroprotection for Premature Infants: Automated Analysis of Neurological Signals for Early Detection of Brain Injury/ info:eu-repo/grantAgreement/SFI/SFI Starting Investigator Research Grant/18/SIRG/5483/IE/A trilogy of stressors in the NICU: Towards therapy for preterm adversity./ © 2024, Minoo Ashoori. https://creativecommons.org/licenses/by-nc-nd/4.0/ Near-infrared spectroscopy (NIRS) Regional cerebral oxygen saturation (rcSO2) Peripheral oxygen saturation (SpO2) Prolonged relative desaturation (PRD) Extreme gradient boosting (XGBoost) Convolutional neural network (CNN) Magnetic resonance imaging (MRI) Hypoxic-ischaemic encephalopathy (HIE) Intraventricular haemorrhage (IVH) Machine learning Deep learning Neonatal brain injury Oxygen delivery AI-based analysis of cerebral oxygenation in preterm and term infants Doctoral thesis Doctoral PhD - Doctor of Philosophyinfo:eu-repo/semantics/openAccess