Exploring the use of machine learning to assess the respiratory function of preterm infants

Thesis

Background: Preterm birth is associated with a number of pathologies that affect respiratory function. The identification of reduced lung function could aid with clinical diagnosis, earlier intervention and improved clinical outcomes for preterm infants. We explored the use of inter-breath intervals to predict the age at which healthy preterm infants’ lungs are functioning. This can provide the groundwork to build a clinical tool that can assess preterm lung function in clinical practice.
Methods: Inter-breath intervals, measured through vital signs monitoring, were longitudinally recorded in healthy preterm infants with a post-menstrual age (PMA) < 37 weeks. Dataset 1, consisting of data from 32 infants, was analysed to compute 49 respiratory and statistical features. The relationship of these features with PMA was assessed through linear regression models. All features were used as inputs to train selected machine learning models to produce a predicted PMA. Mean Absolute Error (MAE) values were used to assess model accuracy.
Machine learning models with higher levels of accuracy were selected for the next stage of analysis. Inter-breath interval data from dataset 2 were analysed, consisting of 66 infants across 161 recordings. 50 features were extracted and used as inputs to train the selected models to output a predicted PMA. The most accurate model was used for further analysis to assess whether performance is affected by sex (paired t-test), ventilation method (ANOVA), and post-natal age (linear regression).
Results: 12 features from dataset 1 and 31 features from dataset 2 had a significant relationship with PMA (p < 0.05). The most accurate model was the bagged trees model trained on all 50 features, with a MAE of 1.30 weeks. Sex (p = 0.17), ventilation method (p = 0.79) and post-natal age (p = 0.99) did not affect model performance. Conclusion: Inter-breath interval data offers novel directions for assessing the respiratory function of preterm infants.

Authors: Ketheeswaranathan, V

• DOI: 10.5287/ora-6rxqrnkxe
• Type of award: MSc by Research
• Level of award: Masters
• Awarding institution: University of Oxford
• Language: English
• Keywords: paediatrics; respiratory; neonatology; machine learning; premature; prediction; preterm; neonates
• Subjects: Pediatrics