Brain imaging analysis for prediction of developmental problems in pre-term infants

Abstract

Amplitude-electroencephalography and Magnetic Resonance Imaging are two non-invasive methods of examining neurological data. Both of these have demonstrated promising capabilities in the prediction of long-term neurodevelopmental outcomes for extremely preterm infants. Therefore, this thesis compared neurodevelopmental outcome predictions made by aEEG-EEG, MRI datasets, and their combination. Further, it tested the application of multiple feature reduction techniques to reduce model complexity. Moreover, it delved into an investigation of the affect caused by various scaling factors on the MRI dataset on the outcome predictions. Results show that with factor analysis employed on the aEEG-EEG dataset and the unscaled version of the MRI dataset, regression models achieved moderate to high performance scores (ranging from r=0.5407 to r=0.9173), while classification models achieved balanced accuracies ranging from 0.795 to 0.907. This thesis provides a basis for further research into multiple modality predictions, hinging on the ability to overcome the hurdle of data shortages.