Cardiac arrest is a life-changing event. In the UK alone there are over 30 000 reported out-of-hospital cardiac arrests every year. In the case of out-of-hospital cardiac arrests over 90% prove to be fatal, and of the ones who survive and are admitted to the intensive care unit, about 84% end up in a comatose state. Of those who end up in a coma, around two thirds will die as a result of hypoxic ischaemic brain injury. To improve the in-hospital critical care of comatose patients, I have developed a model to gauge the predicted patient outcome based on their electroencephalogram (EEG). EEGs do not measure the activity of single neurons, rather when a large group of nearby neurons depolorise in a short time span, the rapid change in ion concentrations ripples through the surrounding medium similarly to a wave and is recorded on an EEG. Additionally, neurons often fire periodically, and modulate their frequency to conduct information, which results in wave-like cyclic patterns on an EEG. Due to this physiology, I have chosen a time-frequency analysis method known as a synchrosqueezing. For considering both time and frequency domain features together, I have used a multi-dimensional convolutional neural network (CNN) to extract features from the resulting time-frequency representation of the EEG. As the goal is to predict the outcome of a patient's neurological recovery, I have used a model utilising positional encoding as well as self-attention to observe relative changes in patient's EEG over time. For training and evaluation, I am usinng training data for the PhysioNet Challenge 2023. I am using 5-fold cross-validation with an attempt to balance each fold in terms of age, sex and patient outcome. I have achieved a 2-class classification accuracy of 63.27% for predicting the patient outcome.